program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-milinternal", ""}, {"coremltools-version", "9.0"}})]
{
    func main<ios18>(tensor<fp32, [1, 128, 1501]> mel, tensor<int32, [1]> mel_length) {
            tensor<int32, [1]> var_2011_axes_0 = const()[name = string("op_2011_axes_0"), val = tensor<int32, [1]>([1])];
            string L_to_fp16_dtype_0 = const()[name = string("L_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1]> mel_length_to_fp16 = cast(dtype = L_to_fp16_dtype_0, x = mel_length)[name = string("cast_8")];
            tensor<fp16, [1, 1]> var_2011_cast_fp16 = expand_dims(axes = var_2011_axes_0, x = mel_length_to_fp16)[name = string("op_2011_cast_fp16")];
            tensor<fp16, [1, 1501]> midx_to_fp16 = const()[name = string("midx_to_fp16"), val = tensor<fp16, [1, 1501]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<bool, [1, 1501]> var_2012_cast_fp16 = less(x = midx_to_fp16, y = var_2011_cast_fp16)[name = string("op_2012_cast_fp16")];
            string attention_mask_dtype_0 = const()[name = string("attention_mask_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_features_perm_0 = const()[name = string("input_features_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string mel_to_fp16_dtype_0 = const()[name = string("mel_to_fp16_dtype_0"), val = string("fp16")];
            int32 var_2025 = const()[name = string("op_2025"), val = int32(2)];
            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 128, 1501]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = string("cast_7")];
            tensor<fp16, [1, 1501, 128]> input_features_cast_fp16 = transpose(perm = input_features_perm_0, x = mel_to_fp16)[name = string("transpose_146")];
            tensor<fp16, [1, 1, 1501, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = input_features_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<int32, [1]> input_lengths_axes_0 = const()[name = string("input_lengths_axes_0"), val = tensor<int32, [1]>([-1])];
            bool input_lengths_keep_dims_0 = const()[name = string("input_lengths_keep_dims_0"), val = bool(false)];
            tensor<int32, [1, 1501]> attention_mask = cast(dtype = attention_mask_dtype_0, x = var_2012_cast_fp16)[name = string("cast_6")];
            tensor<int32, [1]> input_lengths = reduce_sum(axes = input_lengths_axes_0, keep_dims = input_lengths_keep_dims_0, x = attention_mask)[name = string("input_lengths")];
            string hidden_states_1_pad_type_0 = const()[name = string("hidden_states_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_1_pad_0 = const()[name = string("hidden_states_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_1_strides_0 = const()[name = string("hidden_states_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> hidden_states_1_dilations_0 = const()[name = string("hidden_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_1_groups_0 = const()[name = string("hidden_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> e_subsampling_layers_0_weight_to_fp16 = const()[name = string("e_subsampling_layers_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3136)))];
            tensor<fp16, [256]> e_subsampling_layers_0_bias_to_fp16 = const()[name = string("e_subsampling_layers_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7808)))];
            tensor<fp16, [1, 256, 751, 64]> hidden_states_1_cast_fp16 = conv(bias = e_subsampling_layers_0_bias_to_fp16, dilations = hidden_states_1_dilations_0, groups = hidden_states_1_groups_0, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = hidden_states_1_strides_0, weight = e_subsampling_layers_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("hidden_states_1_cast_fp16")];
            int32 var_2054 = const()[name = string("op_2054"), val = int32(1)];
            tensor<int32, [1]> var_2055 = add(x = input_lengths, y = var_2054)[name = string("op_2055")];
            int32 var_2056 = const()[name = string("op_2056"), val = int32(1)];
            tensor<int32, [1]> var_2057 = add(x = var_2055, y = var_2056)[name = string("op_2057")];
            int32 var_2058 = const()[name = string("op_2058"), val = int32(3)];
            tensor<int32, [1]> var_2059 = sub(x = var_2057, y = var_2058)[name = string("op_2059")];
            tensor<int32, [1]> floor_div_0 = floor_div(x = var_2059, y = var_2025)[name = string("floor_div_0")];
            string var_2060_to_fp16_dtype_0 = const()[name = string("op_2060_to_fp16_dtype_0"), val = string("fp16")];
            fp16 var_2061_promoted_to_fp16 = const()[name = string("op_2061_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> floor_div_0_to_fp16 = cast(dtype = var_2060_to_fp16_dtype_0, x = floor_div_0)[name = string("cast_5")];
            tensor<fp16, [1]> current_lengths_1_cast_fp16 = add(x = floor_div_0_to_fp16, y = var_2061_promoted_to_fp16)[name = string("current_lengths_1_cast_fp16")];
            tensor<int32, [1]> var_2066_axes_0 = const()[name = string("op_2066_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_2066_cast_fp16 = expand_dims(axes = var_2066_axes_0, x = current_lengths_1_cast_fp16)[name = string("op_2066_cast_fp16")];
            tensor<fp16, [751]> var_2064_promoted_to_fp16 = const()[name = string("op_2064_promoted_to_fp16"), val = tensor<fp16, [751]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8384)))];
            tensor<bool, [1, 751]> channel_mask_1_cast_fp16 = less(x = var_2064_promoted_to_fp16, y = var_2066_cast_fp16)[name = string("channel_mask_1_cast_fp16")];
            tensor<int32, [1]> var_2069_axes_0 = const()[name = string("op_2069_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 751]> var_2069 = expand_dims(axes = var_2069_axes_0, x = channel_mask_1_cast_fp16)[name = string("op_2069")];
            tensor<int32, [1]> var_2071_axes_0 = const()[name = string("op_2071_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 751, 1]> var_2071 = expand_dims(axes = var_2071_axes_0, x = var_2069)[name = string("op_2071")];
            string var_2071_promoted_to_fp16_dtype_0 = const()[name = string("op_2071_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 751, 1]> var_2071_to_fp16 = cast(dtype = var_2071_promoted_to_fp16_dtype_0, x = var_2071)[name = string("cast_4")];
            tensor<fp16, [1, 256, 751, 64]> input_3_cast_fp16 = mul(x = hidden_states_1_cast_fp16, y = var_2071_to_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(256)];
            tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> e_subsampling_layers_2_weight_to_fp16 = const()[name = string("e_subsampling_layers_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
            tensor<fp16, [256]> e_subsampling_layers_2_bias_to_fp16 = const()[name = string("e_subsampling_layers_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14656)))];
            tensor<fp16, [1, 256, 376, 32]> hidden_states_3_cast_fp16 = conv(bias = e_subsampling_layers_2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = e_subsampling_layers_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("hidden_states_3_cast_fp16")];
            fp16 var_2081_promoted_to_fp16 = const()[name = string("op_2081_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2082_cast_fp16 = add(x = current_lengths_1_cast_fp16, y = var_2081_promoted_to_fp16)[name = string("op_2082_cast_fp16")];
            fp16 var_2083_promoted_to_fp16 = const()[name = string("op_2083_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2084_cast_fp16 = add(x = var_2082_cast_fp16, y = var_2083_promoted_to_fp16)[name = string("op_2084_cast_fp16")];
            fp16 var_2085_promoted_to_fp16 = const()[name = string("op_2085_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_2086_cast_fp16 = sub(x = var_2084_cast_fp16, y = var_2085_promoted_to_fp16)[name = string("op_2086_cast_fp16")];
            fp16 var_2025_promoted_to_fp16 = const()[name = string("op_2025_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_1_cast_fp16 = floor_div(x = var_2086_cast_fp16, y = var_2025_promoted_to_fp16)[name = string("floor_div_1_cast_fp16")];
            fp16 var_2088_promoted_to_fp16 = const()[name = string("op_2088_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> current_lengths_3_cast_fp16 = add(x = floor_div_1_cast_fp16, y = var_2088_promoted_to_fp16)[name = string("current_lengths_3_cast_fp16")];
            tensor<int32, [1]> var_2093_axes_0 = const()[name = string("op_2093_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_2093_cast_fp16 = expand_dims(axes = var_2093_axes_0, x = current_lengths_3_cast_fp16)[name = string("op_2093_cast_fp16")];
            tensor<fp16, [376]> var_2091_promoted_to_fp16 = const()[name = string("op_2091_promoted_to_fp16"), val = tensor<fp16, [376]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15232)))];
            tensor<bool, [1, 376]> channel_mask_3_cast_fp16 = less(x = var_2091_promoted_to_fp16, y = var_2093_cast_fp16)[name = string("channel_mask_3_cast_fp16")];
            tensor<int32, [1]> var_2096_axes_0 = const()[name = string("op_2096_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 376]> var_2096 = expand_dims(axes = var_2096_axes_0, x = channel_mask_3_cast_fp16)[name = string("op_2096")];
            tensor<int32, [1]> var_2098_axes_0 = const()[name = string("op_2098_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 376, 1]> var_2098 = expand_dims(axes = var_2098_axes_0, x = var_2096)[name = string("op_2098")];
            string var_2098_promoted_to_fp16_dtype_0 = const()[name = string("op_2098_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 376, 1]> var_2098_to_fp16 = cast(dtype = var_2098_promoted_to_fp16_dtype_0, x = var_2098)[name = string("cast_3")];
            tensor<fp16, [1, 256, 376, 32]> input_7_cast_fp16 = mul(x = hidden_states_3_cast_fp16, y = var_2098_to_fp16)[name = string("input_7_cast_fp16")];
            string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> e_subsampling_layers_3_weight_to_fp16 = const()[name = string("e_subsampling_layers_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16064)))];
            tensor<fp16, [256]> e_subsampling_layers_3_bias_to_fp16 = const()[name = string("e_subsampling_layers_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147200)))];
            tensor<fp16, [1, 256, 376, 32]> hidden_states_5_cast_fp16 = conv(bias = e_subsampling_layers_3_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = e_subsampling_layers_3_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_9_cast_fp16 = mul(x = hidden_states_5_cast_fp16, y = var_2098_to_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(256)];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> e_subsampling_layers_5_weight_to_fp16 = const()[name = string("e_subsampling_layers_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147776)))];
            tensor<fp16, [256]> e_subsampling_layers_5_bias_to_fp16 = const()[name = string("e_subsampling_layers_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152448)))];
            tensor<fp16, [1, 256, 188, 16]> hidden_states_7_cast_fp16 = conv(bias = e_subsampling_layers_5_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = e_subsampling_layers_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            fp16 var_2125_promoted_to_fp16 = const()[name = string("op_2125_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2126_cast_fp16 = add(x = current_lengths_3_cast_fp16, y = var_2125_promoted_to_fp16)[name = string("op_2126_cast_fp16")];
            fp16 var_2127_promoted_to_fp16 = const()[name = string("op_2127_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2128_cast_fp16 = add(x = var_2126_cast_fp16, y = var_2127_promoted_to_fp16)[name = string("op_2128_cast_fp16")];
            fp16 var_2129_promoted_to_fp16 = const()[name = string("op_2129_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_2130_cast_fp16 = sub(x = var_2128_cast_fp16, y = var_2129_promoted_to_fp16)[name = string("op_2130_cast_fp16")];
            fp16 var_2025_promoted_1_to_fp16 = const()[name = string("op_2025_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_2_cast_fp16 = floor_div(x = var_2130_cast_fp16, y = var_2025_promoted_1_to_fp16)[name = string("floor_div_2_cast_fp16")];
            fp16 var_2132_promoted_to_fp16 = const()[name = string("op_2132_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> current_lengths_cast_fp16 = add(x = floor_div_2_cast_fp16, y = var_2132_promoted_to_fp16)[name = string("current_lengths_cast_fp16")];
            tensor<int32, [1]> var_2137_axes_0 = const()[name = string("op_2137_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_2137_cast_fp16 = expand_dims(axes = var_2137_axes_0, x = current_lengths_cast_fp16)[name = string("op_2137_cast_fp16")];
            tensor<fp16, [188]> var_2135_promoted_to_fp16 = const()[name = string("op_2135_promoted_to_fp16"), val = tensor<fp16, [188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153024)))];
            tensor<bool, [1, 188]> channel_mask_7_cast_fp16 = less(x = var_2135_promoted_to_fp16, y = var_2137_cast_fp16)[name = string("channel_mask_7_cast_fp16")];
            tensor<int32, [1]> var_2140_axes_0 = const()[name = string("op_2140_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188]> var_2140 = expand_dims(axes = var_2140_axes_0, x = channel_mask_7_cast_fp16)[name = string("op_2140")];
            tensor<int32, [1]> var_2142_axes_0 = const()[name = string("op_2142_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<bool, [1, 1, 188, 1]> var_2142 = expand_dims(axes = var_2142_axes_0, x = var_2140)[name = string("op_2142")];
            string var_2142_promoted_to_fp16_dtype_0 = const()[name = string("op_2142_promoted_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, 188, 1]> var_2142_to_fp16 = cast(dtype = var_2142_promoted_to_fp16_dtype_0, x = var_2142)[name = string("cast_2")];
            tensor<fp16, [1, 256, 188, 16]> input_13_cast_fp16 = mul(x = hidden_states_7_cast_fp16, y = var_2142_to_fp16)[name = string("input_13_cast_fp16")];
            string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> e_subsampling_layers_6_weight_to_fp16 = const()[name = string("e_subsampling_layers_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153472)))];
            tensor<fp16, [256]> e_subsampling_layers_6_bias_to_fp16 = const()[name = string("e_subsampling_layers_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284608)))];
            tensor<fp16, [1, 256, 188, 16]> hidden_states_9_cast_fp16 = conv(bias = e_subsampling_layers_6_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = e_subsampling_layers_6_weight_to_fp16, x = input_13_cast_fp16)[name = string("hidden_states_9_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> input_15_cast_fp16 = mul(x = hidden_states_9_cast_fp16, y = var_2142_to_fp16)[name = string("input_15_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> hidden_states_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("hidden_states_cast_fp16")];
            tensor<int32, [4]> var_2162_perm_0 = const()[name = string("op_2162_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2165 = const()[name = string("op_2165"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 256, 16]> var_2162_cast_fp16 = transpose(perm = var_2162_perm_0, x = hidden_states_cast_fp16)[name = string("transpose_145")];
            tensor<fp16, [1, 188, 4096]> input_17_cast_fp16 = reshape(shape = var_2165, x = var_2162_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_subsampling_linear_weight_to_fp16 = const()[name = string("e_subsampling_linear_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285184)))];
            tensor<fp16, [1024]> e_subsampling_linear_bias_to_fp16 = const()[name = string("e_subsampling_linear_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8673856)))];
            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = e_subsampling_linear_bias_to_fp16, weight = e_subsampling_linear_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
            fp16 var_2171_to_fp16 = const()[name = string("op_2171_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2172_cast_fp16 = sub(x = mel_length_to_fp16, y = var_2171_to_fp16)[name = string("op_2172_cast_fp16")];
            fp16 _inversed_2174_y_0_to_fp16 = const()[name = string("_inversed_2174_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_2174_cast_fp16 = mul(x = var_2172_cast_fp16, y = _inversed_2174_y_0_to_fp16)[name = string("_inversed_2174_cast_fp16")];
            fp16 var_2176_to_fp16 = const()[name = string("op_2176_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2177_cast_fp16 = add(x = _inversed_2174_cast_fp16, y = var_2176_to_fp16)[name = string("op_2177_cast_fp16")];
            tensor<fp16, [1]> Lp_1_cast_fp16 = floor(x = var_2177_cast_fp16)[name = string("Lp_1_cast_fp16")];
            fp16 var_2180_to_fp16 = const()[name = string("op_2180_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2181_cast_fp16 = sub(x = Lp_1_cast_fp16, y = var_2180_to_fp16)[name = string("op_2181_cast_fp16")];
            fp16 _inversed_2183_y_0_to_fp16 = const()[name = string("_inversed_2183_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_2183_cast_fp16 = mul(x = var_2181_cast_fp16, y = _inversed_2183_y_0_to_fp16)[name = string("_inversed_2183_cast_fp16")];
            fp16 var_2185_to_fp16 = const()[name = string("op_2185_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2186_cast_fp16 = add(x = _inversed_2183_cast_fp16, y = var_2185_to_fp16)[name = string("op_2186_cast_fp16")];
            tensor<fp16, [1]> Lp_3_cast_fp16 = floor(x = var_2186_cast_fp16)[name = string("Lp_3_cast_fp16")];
            fp16 var_2189_to_fp16 = const()[name = string("op_2189_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2190_cast_fp16 = sub(x = Lp_3_cast_fp16, y = var_2189_to_fp16)[name = string("op_2190_cast_fp16")];
            fp16 _inversed_2192_y_0_to_fp16 = const()[name = string("_inversed_2192_y_0_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1]> _inversed_2192_cast_fp16 = mul(x = var_2190_cast_fp16, y = _inversed_2192_y_0_to_fp16)[name = string("_inversed_2192_cast_fp16")];
            fp16 var_2194_to_fp16 = const()[name = string("op_2194_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_2195_cast_fp16 = add(x = _inversed_2192_cast_fp16, y = var_2194_to_fp16)[name = string("op_2195_cast_fp16")];
            tensor<fp16, [1]> Lp_cast_fp16 = floor(x = var_2195_cast_fp16)[name = string("Lp_cast_fp16")];
            tensor<int32, [1]> var_2203_axes_0 = const()[name = string("op_2203_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1]> var_2203_cast_fp16 = expand_dims(axes = var_2203_axes_0, x = Lp_cast_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [1, 188]> tidx_to_fp16 = const()[name = string("tidx_to_fp16"), val = tensor<fp16, [1, 188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8675968)))];
            tensor<bool, [1, 188]> var_2204_cast_fp16 = less(x = tidx_to_fp16, y = var_2203_cast_fp16)[name = string("op_2204_cast_fp16")];
            tensor<int32, [1]> var_2221_axes_0 = const()[name = string("op_2221_axes_0"), val = tensor<int32, [1]>([2])];
            string valid_to_fp16_dtype_0 = const()[name = string("valid_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 188]> var_2204_cast_fp16_to_fp16 = cast(dtype = valid_to_fp16_dtype_0, x = var_2204_cast_fp16)[name = string("cast_1")];
            tensor<fp16, [1, 188, 1]> var_2221_cast_fp16 = expand_dims(axes = var_2221_axes_0, x = var_2204_cast_fp16_to_fp16)[name = string("op_2221_cast_fp16")];
            tensor<int32, [1]> var_2228_axes_0 = const()[name = string("op_2228_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 188]> var_2228_cast_fp16 = expand_dims(axes = var_2228_axes_0, x = var_2204_cast_fp16_to_fp16)[name = string("op_2228_cast_fp16")];
            tensor<fp16, [1, 188, 188]> var_2234_cast_fp16 = mul(x = var_2221_cast_fp16, y = var_2228_cast_fp16)[name = string("op_2234_cast_fp16")];
            fp16 var_2235_to_fp16 = const()[name = string("op_2235_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 188, 188]> var_2237_cast_fp16 = sub(x = var_2235_to_fp16, y = var_2234_cast_fp16)[name = string("op_2237_cast_fp16")];
            fp16 var_2238_to_fp16 = const()[name = string("op_2238_to_fp16"), val = fp16(-0x1.388p+13)];
            tensor<fp16, [1, 188, 188]> var_2239_cast_fp16 = mul(x = var_2237_cast_fp16, y = var_2238_to_fp16)[name = string("op_2239_cast_fp16")];
            tensor<int32, [1]> bias_15_axes_0 = const()[name = string("bias_15_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 188, 188]> bias_15_cast_fp16 = expand_dims(axes = bias_15_axes_0, x = var_2239_cast_fp16)[name = string("bias_15_cast_fp16")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8676416)))];
            tensor<fp16, [1024]> e_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8678528)))];
            fp16 var_2248_to_fp16 = const()[name = string("op_2248_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = e_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_2248_to_fp16, gamma = e_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_0_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8680640))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9729280))))[name = string("e_layers_0_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9991488)))];
            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_0_feed_forward1_linear1_weight_to_fp16_lut, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_0_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9999744))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11048384))))[name = string("e_layers_0_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11310592)))];
            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_feed_forward1_linear2_weight_to_fp16_lut, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
            fp16 var_2265_to_fp16 = const()[name = string("op_2265_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2266_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_2265_to_fp16)[name = string("op_2266_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_29_cast_fp16 = add(x = linear_0_cast_fp16, y = var_2266_cast_fp16)[name = string("input_29_cast_fp16")];
            tensor<int32, [1]> input_31_axes_0 = const()[name = string("input_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11312704)))];
            tensor<fp16, [1024]> e_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11314816)))];
            fp16 var_2270_to_fp16 = const()[name = string("op_2270_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = e_layers_0_norm_self_att_bias_to_fp16, epsilon = var_2270_to_fp16, gamma = e_layers_0_norm_self_att_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_0_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11316928))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11579136))))[name = string("e_layers_0_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_self_attn_q_proj_weight_to_fp16_lut, x = input_31_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> var_2283 = const()[name = string("op_2283"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2284_cast_fp16 = reshape(shape = var_2283, x = linear_3_cast_fp16)[name = string("op_2284_cast_fp16")];
            tensor<int32, [4]> q_1_perm_0 = const()[name = string("q_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_0_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11644736))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11906944))))[name = string("e_layers_0_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_self_attn_k_proj_weight_to_fp16_lut, x = input_31_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> var_2295 = const()[name = string("op_2295"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2296_cast_fp16 = reshape(shape = var_2295, x = linear_4_cast_fp16)[name = string("op_2296_cast_fp16")];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_0_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11972544))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12234752))))[name = string("e_layers_0_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_self_attn_v_proj_weight_to_fp16_lut, x = input_31_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [4]> var_2307 = const()[name = string("op_2307"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2308_cast_fp16 = reshape(shape = var_2307, x = linear_5_cast_fp16)[name = string("op_2308_cast_fp16")];
            tensor<int32, [4]> v_1_perm_0 = const()[name = string("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_2332_to_fp16 = const()[name = string("op_2332_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12300352)))];
            tensor<fp16, [1, 8, 188, 128]> q_1_cast_fp16 = transpose(perm = q_1_perm_0, x = var_2284_cast_fp16)[name = string("transpose_144")];
            tensor<fp16, [1, 8, 188, 128]> var_2334_cast_fp16 = add(x = q_1_cast_fp16, y = var_2332_to_fp16)[name = string("op_2334_cast_fp16")];
            bool input_35_transpose_x_0 = const()[name = string("input_35_transpose_x_0"), val = bool(false)];
            bool input_35_transpose_y_0 = const()[name = string("input_35_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_1_to_fp16 = const()[name = string("p_1_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12302464)))];
            tensor<fp16, [1, 8, 188, 375]> input_35_cast_fp16 = matmul(transpose_x = input_35_transpose_x_0, transpose_y = input_35_transpose_y_0, x = var_2334_cast_fp16, y = p_1_to_fp16)[name = string("input_35_cast_fp16")];
            tensor<int32, [8]> var_2341_pad_0 = const()[name = string("op_2341_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2341_mode_0 = const()[name = string("op_2341_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_2341_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_2341_mode_0, pad = var_2341_pad_0, x = input_35_cast_fp16)[name = string("op_2341_cast_fp16")];
            tensor<int32, [4]> var_2346 = const()[name = string("op_2346"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_2347_cast_fp16 = reshape(shape = var_2346, x = var_2341_cast_fp16)[name = string("op_2347_cast_fp16")];
            tensor<int32, [4]> var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2347_cast_fp16)[name = string("op_2362_cast_fp16")];
            tensor<int32, [4]> var_2367 = const()[name = string("op_2367"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2368_cast_fp16 = reshape(shape = var_2367, x = var_2362_cast_fp16)[name = string("op_2368_cast_fp16")];
            tensor<int32, [4]> bd_1_begin_0 = const()[name = string("bd_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_1_end_0 = const()[name = string("bd_1_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_1_end_mask_0 = const()[name = string("bd_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_1_cast_fp16 = slice_by_index(begin = bd_1_begin_0, end = bd_1_end_0, end_mask = bd_1_end_mask_0, x = var_2368_cast_fp16)[name = string("bd_1_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_2379_to_fp16 = const()[name = string("op_2379_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13070528)))];
            tensor<fp16, [1, 8, 188, 128]> var_2381_cast_fp16 = add(x = q_1_cast_fp16, y = var_2379_to_fp16)[name = string("op_2381_cast_fp16")];
            bool ac_1_transpose_x_0 = const()[name = string("ac_1_transpose_x_0"), val = bool(false)];
            bool ac_1_transpose_y_0 = const()[name = string("ac_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_48_cast_fp16 = transpose(perm = transpose_48_perm_0, x = var_2296_cast_fp16)[name = string("transpose_143")];
            tensor<fp16, [1, 8, 188, 188]> ac_1_cast_fp16 = matmul(transpose_x = ac_1_transpose_x_0, transpose_y = ac_1_transpose_y_0, x = var_2381_cast_fp16, y = transpose_48_cast_fp16)[name = string("ac_1_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2387_cast_fp16 = add(x = ac_1_cast_fp16, y = bd_1_cast_fp16)[name = string("op_2387_cast_fp16")];
            fp16 var_2388_to_fp16 = const()[name = string("op_2388_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2389_cast_fp16 = mul(x = var_2387_cast_fp16, y = var_2388_to_fp16)[name = string("op_2389_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2391_cast_fp16 = add(x = var_2389_cast_fp16, y = bias_15_cast_fp16)[name = string("op_2391_cast_fp16")];
            int32 var_2392 = const()[name = string("op_2392"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_1_cast_fp16 = softmax(axis = var_2392, x = var_2391_cast_fp16)[name = string("w_1_cast_fp16")];
            bool var_2395_transpose_x_0 = const()[name = string("op_2395_transpose_x_0"), val = bool(false)];
            bool var_2395_transpose_y_0 = const()[name = string("op_2395_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_2308_cast_fp16)[name = string("transpose_142")];
            tensor<fp16, [1, 8, 188, 128]> var_2395_cast_fp16 = matmul(transpose_x = var_2395_transpose_x_0, transpose_y = var_2395_transpose_y_0, x = w_1_cast_fp16, y = v_1_cast_fp16)[name = string("op_2395_cast_fp16")];
            tensor<int32, [4]> var_2398_perm_0 = const()[name = string("op_2398_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2402 = const()[name = string("op_2402"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2398_cast_fp16 = transpose(perm = var_2398_perm_0, x = var_2395_cast_fp16)[name = string("transpose_141")];
            tensor<fp16, [1, 188, 1024]> input_37_cast_fp16 = reshape(shape = var_2402, x = var_2398_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_0_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13072640))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13334848))))[name = string("e_layers_0_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_self_attn_o_proj_weight_to_fp16_lut, x = input_37_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_39_cast_fp16 = add(x = input_29_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<int32, [1]> x_1_axes_0 = const()[name = string("x_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_0_norm_conv_weight_to_fp16 = const()[name = string("e_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13400448)))];
            tensor<fp16, [1024]> e_layers_0_norm_conv_bias_to_fp16 = const()[name = string("e_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13402560)))];
            fp16 var_2410_to_fp16 = const()[name = string("op_2410_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_1_cast_fp16 = layer_norm(axes = x_1_axes_0, beta = e_layers_0_norm_conv_bias_to_fp16, epsilon = var_2410_to_fp16, gamma = e_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_1_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_0_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13404672))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13929024))))[name = string("e_layers_0_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [2048]> var_2424_bias_0_to_fp16 = const()[name = string("op_2424_bias_0_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14060160)))];
            tensor<fp16, [1, 188, 2048]> var_2424_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_0_conv_pointwise_conv1_weight_to_fp16_lut, x = x_1_cast_fp16)[name = string("op_2424_cast_fp16")];
            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_2427_split_num_splits_0 = const()[name = string("op_2427_split_num_splits_0"), val = int32(2)];
            int32 var_2427_split_axis_0 = const()[name = string("op_2427_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = var_2424_cast_fp16)[name = string("transpose_140")];
            tensor<fp16, [1, 1024, 188]> var_2427_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_2427_split_cast_fp16_1 = split(axis = var_2427_split_axis_0, num_splits = var_2427_split_num_splits_0, x = input_41_cast_fp16)[name = string("op_2427_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2427_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2427_split_cast_fp16_1)[name = string("op_2427_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2427_cast_fp16 = mul(x = var_2427_split_cast_fp16_0, y = var_2427_split_1_sigmoid_cast_fp16)[name = string("op_2427_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_43_cast_fp16 = mul(x = var_2427_cast_fp16, y = var_2228_cast_fp16)[name = string("input_43_cast_fp16")];
            string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14064320)))];
            tensor<fp16, [1024]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14082816)))];
            tensor<fp16, [1, 1024, 188]> input_47_cast_fp16 = conv(bias = const_42_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = const_41_to_fp16, x = input_43_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_5_cast_fp16 = silu(x = input_47_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<int32, [3]> var_2467_perm_0 = const()[name = string("op_2467_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_0_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14084928))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14347136))))[name = string("e_layers_0_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_2467_cast_fp16 = transpose(perm = var_2467_perm_0, x = x_5_cast_fp16)[name = string("transpose_139")];
            tensor<fp16, [1, 188, 1024]> var_2469_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_conv_pointwise_conv2_weight_to_fp16_lut, x = var_2467_cast_fp16)[name = string("op_2469_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_49_cast_fp16 = add(x = input_39_cast_fp16, y = var_2469_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<int32, [1]> input_51_axes_0 = const()[name = string("input_51_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14412736)))];
            tensor<fp16, [1024]> e_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14414848)))];
            fp16 var_2477_to_fp16 = const()[name = string("op_2477_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_51_cast_fp16 = layer_norm(axes = input_51_axes_0, beta = e_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_2477_to_fp16, gamma = e_layers_0_norm_feed_forward2_weight_to_fp16, x = input_49_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_0_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14416960))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15465600))))[name = string("e_layers_0_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_0_feed_forward2_linear1_weight_to_fp16_lut, x = input_51_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_55_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_55_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_0_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15727808))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16776448))))[name = string("e_layers_0_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_0_feed_forward2_linear2_weight_to_fp16_lut, x = input_55_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 var_2494_to_fp16 = const()[name = string("op_2494_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2495_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_2494_to_fp16)[name = string("op_2495_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_59_cast_fp16 = add(x = input_49_cast_fp16, y = var_2495_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<int32, [1]> input_61_axes_0 = const()[name = string("input_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_0_norm_out_weight_to_fp16 = const()[name = string("e_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17038656)))];
            tensor<fp16, [1024]> e_layers_0_norm_out_bias_to_fp16 = const()[name = string("e_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17040768)))];
            fp16 var_2499_to_fp16 = const()[name = string("op_2499_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = e_layers_0_norm_out_bias_to_fp16, epsilon = var_2499_to_fp16, gamma = e_layers_0_norm_out_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<int32, [1]> input_63_axes_0 = const()[name = string("input_63_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17042880)))];
            tensor<fp16, [1024]> e_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17044992)))];
            fp16 var_2506_to_fp16 = const()[name = string("op_2506_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = e_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_2506_to_fp16, gamma = e_layers_1_norm_feed_forward1_weight_to_fp16, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_1_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17047104))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18095744))))[name = string("e_layers_1_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_1_feed_forward1_linear1_weight_to_fp16_lut, x = input_63_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_67_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_1_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18357952))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19406592))))[name = string("e_layers_1_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_feed_forward1_linear2_weight_to_fp16_lut, x = input_67_cast_fp16)[name = string("linear_11_cast_fp16")];
            fp16 var_2523_to_fp16 = const()[name = string("op_2523_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2524_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_2523_to_fp16)[name = string("op_2524_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_71_cast_fp16 = add(x = input_61_cast_fp16, y = var_2524_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19668800)))];
            tensor<fp16, [1024]> e_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19670912)))];
            fp16 var_2528_to_fp16 = const()[name = string("op_2528_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = e_layers_1_norm_self_att_bias_to_fp16, epsilon = var_2528_to_fp16, gamma = e_layers_1_norm_self_att_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_1_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19673024))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19935232))))[name = string("e_layers_1_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_self_attn_q_proj_weight_to_fp16_lut, x = input_73_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> var_2541 = const()[name = string("op_2541"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2542_cast_fp16 = reshape(shape = var_2541, x = linear_12_cast_fp16)[name = string("op_2542_cast_fp16")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = string("q_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_1_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20000832))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20263040))))[name = string("e_layers_1_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_self_attn_k_proj_weight_to_fp16_lut, x = input_73_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> var_2553 = const()[name = string("op_2553"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2554_cast_fp16 = reshape(shape = var_2553, x = linear_13_cast_fp16)[name = string("op_2554_cast_fp16")];
            tensor<int32, [4]> transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_1_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20328640))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20590848))))[name = string("e_layers_1_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_self_attn_v_proj_weight_to_fp16_lut, x = input_73_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> var_2565 = const()[name = string("op_2565"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2566_cast_fp16 = reshape(shape = var_2565, x = linear_14_cast_fp16)[name = string("op_2566_cast_fp16")];
            tensor<int32, [4]> v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_2590_to_fp16 = const()[name = string("op_2590_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20656448)))];
            tensor<fp16, [1, 8, 188, 128]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = var_2542_cast_fp16)[name = string("transpose_138")];
            tensor<fp16, [1, 8, 188, 128]> var_2592_cast_fp16 = add(x = q_3_cast_fp16, y = var_2590_to_fp16)[name = string("op_2592_cast_fp16")];
            bool input_75_transpose_x_0 = const()[name = string("input_75_transpose_x_0"), val = bool(false)];
            bool input_75_transpose_y_0 = const()[name = string("input_75_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_3_to_fp16 = const()[name = string("p_3_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20658560)))];
            tensor<fp16, [1, 8, 188, 375]> input_75_cast_fp16 = matmul(transpose_x = input_75_transpose_x_0, transpose_y = input_75_transpose_y_0, x = var_2592_cast_fp16, y = p_3_to_fp16)[name = string("input_75_cast_fp16")];
            tensor<int32, [8]> var_2599_pad_0 = const()[name = string("op_2599_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2599_mode_0 = const()[name = string("op_2599_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_2599_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_2599_mode_0, pad = var_2599_pad_0, x = input_75_cast_fp16)[name = string("op_2599_cast_fp16")];
            tensor<int32, [4]> var_2604 = const()[name = string("op_2604"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_2605_cast_fp16 = reshape(shape = var_2604, x = var_2599_cast_fp16)[name = string("op_2605_cast_fp16")];
            tensor<int32, [4]> var_2620_begin_0 = const()[name = string("op_2620_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2620_end_0 = const()[name = string("op_2620_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2620_end_mask_0 = const()[name = string("op_2620_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2620_cast_fp16 = slice_by_index(begin = var_2620_begin_0, end = var_2620_end_0, end_mask = var_2620_end_mask_0, x = var_2605_cast_fp16)[name = string("op_2620_cast_fp16")];
            tensor<int32, [4]> var_2625 = const()[name = string("op_2625"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2626_cast_fp16 = reshape(shape = var_2625, x = var_2620_cast_fp16)[name = string("op_2626_cast_fp16")];
            tensor<int32, [4]> bd_3_begin_0 = const()[name = string("bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_3_end_0 = const()[name = string("bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_3_end_mask_0 = const()[name = string("bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_3_cast_fp16 = slice_by_index(begin = bd_3_begin_0, end = bd_3_end_0, end_mask = bd_3_end_mask_0, x = var_2626_cast_fp16)[name = string("bd_3_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21426624)))];
            tensor<fp16, [1, 8, 188, 128]> var_2639_cast_fp16 = add(x = q_3_cast_fp16, y = var_2637_to_fp16)[name = string("op_2639_cast_fp16")];
            bool ac_3_transpose_x_0 = const()[name = string("ac_3_transpose_x_0"), val = bool(false)];
            bool ac_3_transpose_y_0 = const()[name = string("ac_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_49_cast_fp16 = transpose(perm = transpose_49_perm_0, x = var_2554_cast_fp16)[name = string("transpose_137")];
            tensor<fp16, [1, 8, 188, 188]> ac_3_cast_fp16 = matmul(transpose_x = ac_3_transpose_x_0, transpose_y = ac_3_transpose_y_0, x = var_2639_cast_fp16, y = transpose_49_cast_fp16)[name = string("ac_3_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2645_cast_fp16 = add(x = ac_3_cast_fp16, y = bd_3_cast_fp16)[name = string("op_2645_cast_fp16")];
            fp16 var_2646_to_fp16 = const()[name = string("op_2646_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2647_cast_fp16 = mul(x = var_2645_cast_fp16, y = var_2646_to_fp16)[name = string("op_2647_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2649_cast_fp16 = add(x = var_2647_cast_fp16, y = bias_15_cast_fp16)[name = string("op_2649_cast_fp16")];
            int32 var_2650 = const()[name = string("op_2650"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_3_cast_fp16 = softmax(axis = var_2650, x = var_2649_cast_fp16)[name = string("w_3_cast_fp16")];
            bool var_2653_transpose_x_0 = const()[name = string("op_2653_transpose_x_0"), val = bool(false)];
            bool var_2653_transpose_y_0 = const()[name = string("op_2653_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_2566_cast_fp16)[name = string("transpose_136")];
            tensor<fp16, [1, 8, 188, 128]> var_2653_cast_fp16 = matmul(transpose_x = var_2653_transpose_x_0, transpose_y = var_2653_transpose_y_0, x = w_3_cast_fp16, y = v_3_cast_fp16)[name = string("op_2653_cast_fp16")];
            tensor<int32, [4]> var_2656_perm_0 = const()[name = string("op_2656_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2660 = const()[name = string("op_2660"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2656_cast_fp16 = transpose(perm = var_2656_perm_0, x = var_2653_cast_fp16)[name = string("transpose_135")];
            tensor<fp16, [1, 188, 1024]> input_77_cast_fp16 = reshape(shape = var_2660, x = var_2656_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_1_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21428736))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21690944))))[name = string("e_layers_1_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_self_attn_o_proj_weight_to_fp16_lut, x = input_77_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_79_cast_fp16 = add(x = input_71_cast_fp16, y = linear_16_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<int32, [1]> x_7_axes_0 = const()[name = string("x_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_1_norm_conv_weight_to_fp16 = const()[name = string("e_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21756544)))];
            tensor<fp16, [1024]> e_layers_1_norm_conv_bias_to_fp16 = const()[name = string("e_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21758656)))];
            fp16 var_2668_to_fp16 = const()[name = string("op_2668_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_7_cast_fp16 = layer_norm(axes = x_7_axes_0, beta = e_layers_1_norm_conv_bias_to_fp16, epsilon = var_2668_to_fp16, gamma = e_layers_1_norm_conv_weight_to_fp16, x = input_79_cast_fp16)[name = string("x_7_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_1_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21760768))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22285120))))[name = string("e_layers_1_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_2682_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_1_conv_pointwise_conv1_weight_to_fp16_lut, x = x_7_cast_fp16)[name = string("op_2682_cast_fp16")];
            tensor<int32, [3]> input_81_perm_0 = const()[name = string("input_81_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_2685_split_num_splits_0 = const()[name = string("op_2685_split_num_splits_0"), val = int32(2)];
            int32 var_2685_split_axis_0 = const()[name = string("op_2685_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_81_cast_fp16 = transpose(perm = input_81_perm_0, x = var_2682_cast_fp16)[name = string("transpose_134")];
            tensor<fp16, [1, 1024, 188]> var_2685_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_2685_split_cast_fp16_1 = split(axis = var_2685_split_axis_0, num_splits = var_2685_split_num_splits_0, x = input_81_cast_fp16)[name = string("op_2685_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2685_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2685_split_cast_fp16_1)[name = string("op_2685_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2685_cast_fp16 = mul(x = var_2685_split_cast_fp16_0, y = var_2685_split_1_sigmoid_cast_fp16)[name = string("op_2685_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_83_cast_fp16 = mul(x = var_2685_cast_fp16, y = var_2228_cast_fp16)[name = string("input_83_cast_fp16")];
            string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_43_to_fp16 = const()[name = string("const_43_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22416256)))];
            tensor<fp16, [1024]> const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22434752)))];
            tensor<fp16, [1, 1024, 188]> input_87_cast_fp16 = conv(bias = const_44_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = const_43_to_fp16, x = input_83_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_11_cast_fp16 = silu(x = input_87_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [3]> var_2725_perm_0 = const()[name = string("op_2725_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_1_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22436864))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22699072))))[name = string("e_layers_1_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_2725_cast_fp16 = transpose(perm = var_2725_perm_0, x = x_11_cast_fp16)[name = string("transpose_133")];
            tensor<fp16, [1, 188, 1024]> var_2727_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_conv_pointwise_conv2_weight_to_fp16_lut, x = var_2725_cast_fp16)[name = string("op_2727_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_89_cast_fp16 = add(x = input_79_cast_fp16, y = var_2727_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<int32, [1]> input_91_axes_0 = const()[name = string("input_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22764672)))];
            tensor<fp16, [1024]> e_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22766784)))];
            fp16 var_2735_to_fp16 = const()[name = string("op_2735_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_91_cast_fp16 = layer_norm(axes = input_91_axes_0, beta = e_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_2735_to_fp16, gamma = e_layers_1_norm_feed_forward2_weight_to_fp16, x = input_89_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_1_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22768896))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23817536))))[name = string("e_layers_1_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_1_feed_forward2_linear1_weight_to_fp16_lut, x = input_91_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_95_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_1_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24079744))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25128384))))[name = string("e_layers_1_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_1_feed_forward2_linear2_weight_to_fp16_lut, x = input_95_cast_fp16)[name = string("linear_18_cast_fp16")];
            fp16 var_2752_to_fp16 = const()[name = string("op_2752_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2753_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_2752_to_fp16)[name = string("op_2753_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_99_cast_fp16 = add(x = input_89_cast_fp16, y = var_2753_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [1]> input_101_axes_0 = const()[name = string("input_101_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_1_norm_out_weight_to_fp16 = const()[name = string("e_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25390592)))];
            tensor<fp16, [1024]> e_layers_1_norm_out_bias_to_fp16 = const()[name = string("e_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25392704)))];
            fp16 var_2757_to_fp16 = const()[name = string("op_2757_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = e_layers_1_norm_out_bias_to_fp16, epsilon = var_2757_to_fp16, gamma = e_layers_1_norm_out_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<int32, [1]> input_103_axes_0 = const()[name = string("input_103_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25394816)))];
            tensor<fp16, [1024]> e_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25396928)))];
            fp16 var_2764_to_fp16 = const()[name = string("op_2764_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_103_cast_fp16 = layer_norm(axes = input_103_axes_0, beta = e_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_2764_to_fp16, gamma = e_layers_2_norm_feed_forward1_weight_to_fp16, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_2_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25399040))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26447680))))[name = string("e_layers_2_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_2_feed_forward1_linear1_weight_to_fp16_lut, x = input_103_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_107_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_107_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_2_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26709888))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27758528))))[name = string("e_layers_2_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_feed_forward1_linear2_weight_to_fp16_lut, x = input_107_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 var_2781_to_fp16 = const()[name = string("op_2781_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2782_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_2781_to_fp16)[name = string("op_2782_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_111_cast_fp16 = add(x = input_101_cast_fp16, y = var_2782_cast_fp16)[name = string("input_111_cast_fp16")];
            tensor<int32, [1]> input_113_axes_0 = const()[name = string("input_113_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28020736)))];
            tensor<fp16, [1024]> e_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28022848)))];
            fp16 var_2786_to_fp16 = const()[name = string("op_2786_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = e_layers_2_norm_self_att_bias_to_fp16, epsilon = var_2786_to_fp16, gamma = e_layers_2_norm_self_att_weight_to_fp16, x = input_111_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_2_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28024960))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28287168))))[name = string("e_layers_2_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_self_attn_q_proj_weight_to_fp16_lut, x = input_113_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> var_2799 = const()[name = string("op_2799"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2800_cast_fp16 = reshape(shape = var_2799, x = linear_21_cast_fp16)[name = string("op_2800_cast_fp16")];
            tensor<int32, [4]> q_5_perm_0 = const()[name = string("q_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_2_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28352768))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28614976))))[name = string("e_layers_2_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_self_attn_k_proj_weight_to_fp16_lut, x = input_113_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> var_2811 = const()[name = string("op_2811"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2812_cast_fp16 = reshape(shape = var_2811, x = linear_22_cast_fp16)[name = string("op_2812_cast_fp16")];
            tensor<int32, [4]> transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_2_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28680576))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28942784))))[name = string("e_layers_2_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_self_attn_v_proj_weight_to_fp16_lut, x = input_113_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> var_2823 = const()[name = string("op_2823"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_2824_cast_fp16 = reshape(shape = var_2823, x = linear_23_cast_fp16)[name = string("op_2824_cast_fp16")];
            tensor<int32, [4]> v_5_perm_0 = const()[name = string("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_2848_to_fp16 = const()[name = string("op_2848_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29008384)))];
            tensor<fp16, [1, 8, 188, 128]> q_5_cast_fp16 = transpose(perm = q_5_perm_0, x = var_2800_cast_fp16)[name = string("transpose_132")];
            tensor<fp16, [1, 8, 188, 128]> var_2850_cast_fp16 = add(x = q_5_cast_fp16, y = var_2848_to_fp16)[name = string("op_2850_cast_fp16")];
            bool input_115_transpose_x_0 = const()[name = string("input_115_transpose_x_0"), val = bool(false)];
            bool input_115_transpose_y_0 = const()[name = string("input_115_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_5_to_fp16 = const()[name = string("p_5_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29010496)))];
            tensor<fp16, [1, 8, 188, 375]> input_115_cast_fp16 = matmul(transpose_x = input_115_transpose_x_0, transpose_y = input_115_transpose_y_0, x = var_2850_cast_fp16, y = p_5_to_fp16)[name = string("input_115_cast_fp16")];
            tensor<int32, [8]> var_2857_pad_0 = const()[name = string("op_2857_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_2857_mode_0 = const()[name = string("op_2857_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_2857_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_2857_mode_0, pad = var_2857_pad_0, x = input_115_cast_fp16)[name = string("op_2857_cast_fp16")];
            tensor<int32, [4]> var_2862 = const()[name = string("op_2862"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_2863_cast_fp16 = reshape(shape = var_2862, x = var_2857_cast_fp16)[name = string("op_2863_cast_fp16")];
            tensor<int32, [4]> var_2878_begin_0 = const()[name = string("op_2878_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2878_end_0 = const()[name = string("op_2878_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2878_end_mask_0 = const()[name = string("op_2878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2878_cast_fp16 = slice_by_index(begin = var_2878_begin_0, end = var_2878_end_0, end_mask = var_2878_end_mask_0, x = var_2863_cast_fp16)[name = string("op_2878_cast_fp16")];
            tensor<int32, [4]> var_2883 = const()[name = string("op_2883"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_2884_cast_fp16 = reshape(shape = var_2883, x = var_2878_cast_fp16)[name = string("op_2884_cast_fp16")];
            tensor<int32, [4]> bd_5_begin_0 = const()[name = string("bd_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_5_end_0 = const()[name = string("bd_5_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_5_end_mask_0 = const()[name = string("bd_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_5_cast_fp16 = slice_by_index(begin = bd_5_begin_0, end = bd_5_end_0, end_mask = bd_5_end_mask_0, x = var_2884_cast_fp16)[name = string("bd_5_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_2895_to_fp16 = const()[name = string("op_2895_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29778560)))];
            tensor<fp16, [1, 8, 188, 128]> var_2897_cast_fp16 = add(x = q_5_cast_fp16, y = var_2895_to_fp16)[name = string("op_2897_cast_fp16")];
            bool ac_5_transpose_x_0 = const()[name = string("ac_5_transpose_x_0"), val = bool(false)];
            bool ac_5_transpose_y_0 = const()[name = string("ac_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_50_cast_fp16 = transpose(perm = transpose_50_perm_0, x = var_2812_cast_fp16)[name = string("transpose_131")];
            tensor<fp16, [1, 8, 188, 188]> ac_5_cast_fp16 = matmul(transpose_x = ac_5_transpose_x_0, transpose_y = ac_5_transpose_y_0, x = var_2897_cast_fp16, y = transpose_50_cast_fp16)[name = string("ac_5_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2903_cast_fp16 = add(x = ac_5_cast_fp16, y = bd_5_cast_fp16)[name = string("op_2903_cast_fp16")];
            fp16 var_2904_to_fp16 = const()[name = string("op_2904_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_2905_cast_fp16 = mul(x = var_2903_cast_fp16, y = var_2904_to_fp16)[name = string("op_2905_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2907_cast_fp16 = add(x = var_2905_cast_fp16, y = bias_15_cast_fp16)[name = string("op_2907_cast_fp16")];
            int32 var_2908 = const()[name = string("op_2908"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_5_cast_fp16 = softmax(axis = var_2908, x = var_2907_cast_fp16)[name = string("w_5_cast_fp16")];
            bool var_2911_transpose_x_0 = const()[name = string("op_2911_transpose_x_0"), val = bool(false)];
            bool var_2911_transpose_y_0 = const()[name = string("op_2911_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_2824_cast_fp16)[name = string("transpose_130")];
            tensor<fp16, [1, 8, 188, 128]> var_2911_cast_fp16 = matmul(transpose_x = var_2911_transpose_x_0, transpose_y = var_2911_transpose_y_0, x = w_5_cast_fp16, y = v_5_cast_fp16)[name = string("op_2911_cast_fp16")];
            tensor<int32, [4]> var_2914_perm_0 = const()[name = string("op_2914_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2918 = const()[name = string("op_2918"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2914_cast_fp16 = transpose(perm = var_2914_perm_0, x = var_2911_cast_fp16)[name = string("transpose_129")];
            tensor<fp16, [1, 188, 1024]> input_117_cast_fp16 = reshape(shape = var_2918, x = var_2914_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_2_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29780672))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30042880))))[name = string("e_layers_2_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_self_attn_o_proj_weight_to_fp16_lut, x = input_117_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_119_cast_fp16 = add(x = input_111_cast_fp16, y = linear_25_cast_fp16)[name = string("input_119_cast_fp16")];
            tensor<int32, [1]> x_13_axes_0 = const()[name = string("x_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_2_norm_conv_weight_to_fp16 = const()[name = string("e_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30108480)))];
            tensor<fp16, [1024]> e_layers_2_norm_conv_bias_to_fp16 = const()[name = string("e_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30110592)))];
            fp16 var_2926_to_fp16 = const()[name = string("op_2926_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_13_cast_fp16 = layer_norm(axes = x_13_axes_0, beta = e_layers_2_norm_conv_bias_to_fp16, epsilon = var_2926_to_fp16, gamma = e_layers_2_norm_conv_weight_to_fp16, x = input_119_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_2_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30112704))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30637056))))[name = string("e_layers_2_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_2940_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_2_conv_pointwise_conv1_weight_to_fp16_lut, x = x_13_cast_fp16)[name = string("op_2940_cast_fp16")];
            tensor<int32, [3]> input_121_perm_0 = const()[name = string("input_121_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_2943_split_num_splits_0 = const()[name = string("op_2943_split_num_splits_0"), val = int32(2)];
            int32 var_2943_split_axis_0 = const()[name = string("op_2943_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_121_cast_fp16 = transpose(perm = input_121_perm_0, x = var_2940_cast_fp16)[name = string("transpose_128")];
            tensor<fp16, [1, 1024, 188]> var_2943_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_2943_split_cast_fp16_1 = split(axis = var_2943_split_axis_0, num_splits = var_2943_split_num_splits_0, x = input_121_cast_fp16)[name = string("op_2943_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2943_split_1_sigmoid_cast_fp16 = sigmoid(x = var_2943_split_cast_fp16_1)[name = string("op_2943_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_2943_cast_fp16 = mul(x = var_2943_split_cast_fp16_0, y = var_2943_split_1_sigmoid_cast_fp16)[name = string("op_2943_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_123_cast_fp16 = mul(x = var_2943_cast_fp16, y = var_2228_cast_fp16)[name = string("input_123_cast_fp16")];
            string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30768192)))];
            tensor<fp16, [1024]> const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30786688)))];
            tensor<fp16, [1, 1024, 188]> input_127_cast_fp16 = conv(bias = const_46_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = const_45_to_fp16, x = input_123_cast_fp16)[name = string("input_127_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_17_cast_fp16 = silu(x = input_127_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [3]> var_2983_perm_0 = const()[name = string("op_2983_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_2_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30788800))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31051008))))[name = string("e_layers_2_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_2983_cast_fp16 = transpose(perm = var_2983_perm_0, x = x_17_cast_fp16)[name = string("transpose_127")];
            tensor<fp16, [1, 188, 1024]> var_2985_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_conv_pointwise_conv2_weight_to_fp16_lut, x = var_2983_cast_fp16)[name = string("op_2985_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_129_cast_fp16 = add(x = input_119_cast_fp16, y = var_2985_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<int32, [1]> input_131_axes_0 = const()[name = string("input_131_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31116608)))];
            tensor<fp16, [1024]> e_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31118720)))];
            fp16 var_2993_to_fp16 = const()[name = string("op_2993_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_131_cast_fp16 = layer_norm(axes = input_131_axes_0, beta = e_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_2993_to_fp16, gamma = e_layers_2_norm_feed_forward2_weight_to_fp16, x = input_129_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_2_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31120832))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32169472))))[name = string("e_layers_2_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_2_feed_forward2_linear1_weight_to_fp16_lut, x = input_131_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_135_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_135_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_2_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32431680))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33480320))))[name = string("e_layers_2_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_2_feed_forward2_linear2_weight_to_fp16_lut, x = input_135_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 var_3010_to_fp16 = const()[name = string("op_3010_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3011_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_3010_to_fp16)[name = string("op_3011_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_139_cast_fp16 = add(x = input_129_cast_fp16, y = var_3011_cast_fp16)[name = string("input_139_cast_fp16")];
            tensor<int32, [1]> input_141_axes_0 = const()[name = string("input_141_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_2_norm_out_weight_to_fp16 = const()[name = string("e_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33742528)))];
            tensor<fp16, [1024]> e_layers_2_norm_out_bias_to_fp16 = const()[name = string("e_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33744640)))];
            fp16 var_3015_to_fp16 = const()[name = string("op_3015_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = e_layers_2_norm_out_bias_to_fp16, epsilon = var_3015_to_fp16, gamma = e_layers_2_norm_out_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<int32, [1]> input_143_axes_0 = const()[name = string("input_143_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33746752)))];
            tensor<fp16, [1024]> e_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33748864)))];
            fp16 var_3022_to_fp16 = const()[name = string("op_3022_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_143_cast_fp16 = layer_norm(axes = input_143_axes_0, beta = e_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_3022_to_fp16, gamma = e_layers_3_norm_feed_forward1_weight_to_fp16, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_3_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33750976))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34799616))))[name = string("e_layers_3_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_3_feed_forward1_linear1_weight_to_fp16_lut, x = input_143_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_147_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_147_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_3_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35061824))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36110464))))[name = string("e_layers_3_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_feed_forward1_linear2_weight_to_fp16_lut, x = input_147_cast_fp16)[name = string("linear_29_cast_fp16")];
            fp16 var_3039_to_fp16 = const()[name = string("op_3039_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3040_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_3039_to_fp16)[name = string("op_3040_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_151_cast_fp16 = add(x = input_141_cast_fp16, y = var_3040_cast_fp16)[name = string("input_151_cast_fp16")];
            tensor<int32, [1]> input_153_axes_0 = const()[name = string("input_153_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36372672)))];
            tensor<fp16, [1024]> e_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36374784)))];
            fp16 var_3044_to_fp16 = const()[name = string("op_3044_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_153_cast_fp16 = layer_norm(axes = input_153_axes_0, beta = e_layers_3_norm_self_att_bias_to_fp16, epsilon = var_3044_to_fp16, gamma = e_layers_3_norm_self_att_weight_to_fp16, x = input_151_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_3_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36376896))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36639104))))[name = string("e_layers_3_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_self_attn_q_proj_weight_to_fp16_lut, x = input_153_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_3057 = const()[name = string("op_3057"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3058_cast_fp16 = reshape(shape = var_3057, x = linear_30_cast_fp16)[name = string("op_3058_cast_fp16")];
            tensor<int32, [4]> q_7_perm_0 = const()[name = string("q_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_3_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36704704))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36966912))))[name = string("e_layers_3_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_self_attn_k_proj_weight_to_fp16_lut, x = input_153_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> var_3069 = const()[name = string("op_3069"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3070_cast_fp16 = reshape(shape = var_3069, x = linear_31_cast_fp16)[name = string("op_3070_cast_fp16")];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_3_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37032512))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37294720))))[name = string("e_layers_3_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_self_attn_v_proj_weight_to_fp16_lut, x = input_153_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [4]> var_3081 = const()[name = string("op_3081"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3082_cast_fp16 = reshape(shape = var_3081, x = linear_32_cast_fp16)[name = string("op_3082_cast_fp16")];
            tensor<int32, [4]> v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_3106_to_fp16 = const()[name = string("op_3106_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37360320)))];
            tensor<fp16, [1, 8, 188, 128]> q_7_cast_fp16 = transpose(perm = q_7_perm_0, x = var_3058_cast_fp16)[name = string("transpose_126")];
            tensor<fp16, [1, 8, 188, 128]> var_3108_cast_fp16 = add(x = q_7_cast_fp16, y = var_3106_to_fp16)[name = string("op_3108_cast_fp16")];
            bool input_155_transpose_x_0 = const()[name = string("input_155_transpose_x_0"), val = bool(false)];
            bool input_155_transpose_y_0 = const()[name = string("input_155_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_7_to_fp16 = const()[name = string("p_7_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37362432)))];
            tensor<fp16, [1, 8, 188, 375]> input_155_cast_fp16 = matmul(transpose_x = input_155_transpose_x_0, transpose_y = input_155_transpose_y_0, x = var_3108_cast_fp16, y = p_7_to_fp16)[name = string("input_155_cast_fp16")];
            tensor<int32, [8]> var_3115_pad_0 = const()[name = string("op_3115_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3115_mode_0 = const()[name = string("op_3115_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3115_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_3115_mode_0, pad = var_3115_pad_0, x = input_155_cast_fp16)[name = string("op_3115_cast_fp16")];
            tensor<int32, [4]> var_3120 = const()[name = string("op_3120"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_3121_cast_fp16 = reshape(shape = var_3120, x = var_3115_cast_fp16)[name = string("op_3121_cast_fp16")];
            tensor<int32, [4]> var_3136_begin_0 = const()[name = string("op_3136_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3136_end_0 = const()[name = string("op_3136_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3136_end_mask_0 = const()[name = string("op_3136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3136_cast_fp16 = slice_by_index(begin = var_3136_begin_0, end = var_3136_end_0, end_mask = var_3136_end_mask_0, x = var_3121_cast_fp16)[name = string("op_3136_cast_fp16")];
            tensor<int32, [4]> var_3141 = const()[name = string("op_3141"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3142_cast_fp16 = reshape(shape = var_3141, x = var_3136_cast_fp16)[name = string("op_3142_cast_fp16")];
            tensor<int32, [4]> bd_7_begin_0 = const()[name = string("bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_7_end_0 = const()[name = string("bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_7_end_mask_0 = const()[name = string("bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_7_cast_fp16 = slice_by_index(begin = bd_7_begin_0, end = bd_7_end_0, end_mask = bd_7_end_mask_0, x = var_3142_cast_fp16)[name = string("bd_7_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_3153_to_fp16 = const()[name = string("op_3153_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38130496)))];
            tensor<fp16, [1, 8, 188, 128]> var_3155_cast_fp16 = add(x = q_7_cast_fp16, y = var_3153_to_fp16)[name = string("op_3155_cast_fp16")];
            bool ac_7_transpose_x_0 = const()[name = string("ac_7_transpose_x_0"), val = bool(false)];
            bool ac_7_transpose_y_0 = const()[name = string("ac_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_51_cast_fp16 = transpose(perm = transpose_51_perm_0, x = var_3070_cast_fp16)[name = string("transpose_125")];
            tensor<fp16, [1, 8, 188, 188]> ac_7_cast_fp16 = matmul(transpose_x = ac_7_transpose_x_0, transpose_y = ac_7_transpose_y_0, x = var_3155_cast_fp16, y = transpose_51_cast_fp16)[name = string("ac_7_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3161_cast_fp16 = add(x = ac_7_cast_fp16, y = bd_7_cast_fp16)[name = string("op_3161_cast_fp16")];
            fp16 var_3162_to_fp16 = const()[name = string("op_3162_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3163_cast_fp16 = mul(x = var_3161_cast_fp16, y = var_3162_to_fp16)[name = string("op_3163_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3165_cast_fp16 = add(x = var_3163_cast_fp16, y = bias_15_cast_fp16)[name = string("op_3165_cast_fp16")];
            int32 var_3166 = const()[name = string("op_3166"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_7_cast_fp16 = softmax(axis = var_3166, x = var_3165_cast_fp16)[name = string("w_7_cast_fp16")];
            bool var_3169_transpose_x_0 = const()[name = string("op_3169_transpose_x_0"), val = bool(false)];
            bool var_3169_transpose_y_0 = const()[name = string("op_3169_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_3082_cast_fp16)[name = string("transpose_124")];
            tensor<fp16, [1, 8, 188, 128]> var_3169_cast_fp16 = matmul(transpose_x = var_3169_transpose_x_0, transpose_y = var_3169_transpose_y_0, x = w_7_cast_fp16, y = v_7_cast_fp16)[name = string("op_3169_cast_fp16")];
            tensor<int32, [4]> var_3172_perm_0 = const()[name = string("op_3172_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3176 = const()[name = string("op_3176"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3172_cast_fp16 = transpose(perm = var_3172_perm_0, x = var_3169_cast_fp16)[name = string("transpose_123")];
            tensor<fp16, [1, 188, 1024]> input_157_cast_fp16 = reshape(shape = var_3176, x = var_3172_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_3_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38132608))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38394816))))[name = string("e_layers_3_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_self_attn_o_proj_weight_to_fp16_lut, x = input_157_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_159_cast_fp16 = add(x = input_151_cast_fp16, y = linear_34_cast_fp16)[name = string("input_159_cast_fp16")];
            tensor<int32, [1]> x_19_axes_0 = const()[name = string("x_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_3_norm_conv_weight_to_fp16 = const()[name = string("e_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38460416)))];
            tensor<fp16, [1024]> e_layers_3_norm_conv_bias_to_fp16 = const()[name = string("e_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38462528)))];
            fp16 var_3184_to_fp16 = const()[name = string("op_3184_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_19_cast_fp16 = layer_norm(axes = x_19_axes_0, beta = e_layers_3_norm_conv_bias_to_fp16, epsilon = var_3184_to_fp16, gamma = e_layers_3_norm_conv_weight_to_fp16, x = input_159_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_3_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38464640))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38988992))))[name = string("e_layers_3_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_3198_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_3_conv_pointwise_conv1_weight_to_fp16_lut, x = x_19_cast_fp16)[name = string("op_3198_cast_fp16")];
            tensor<int32, [3]> input_161_perm_0 = const()[name = string("input_161_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_3201_split_num_splits_0 = const()[name = string("op_3201_split_num_splits_0"), val = int32(2)];
            int32 var_3201_split_axis_0 = const()[name = string("op_3201_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_161_cast_fp16 = transpose(perm = input_161_perm_0, x = var_3198_cast_fp16)[name = string("transpose_122")];
            tensor<fp16, [1, 1024, 188]> var_3201_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_3201_split_cast_fp16_1 = split(axis = var_3201_split_axis_0, num_splits = var_3201_split_num_splits_0, x = input_161_cast_fp16)[name = string("op_3201_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3201_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3201_split_cast_fp16_1)[name = string("op_3201_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3201_cast_fp16 = mul(x = var_3201_split_cast_fp16_0, y = var_3201_split_1_sigmoid_cast_fp16)[name = string("op_3201_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_163_cast_fp16 = mul(x = var_3201_cast_fp16, y = var_2228_cast_fp16)[name = string("input_163_cast_fp16")];
            string input_165_pad_type_0 = const()[name = string("input_165_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_165_groups_0 = const()[name = string("input_165_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_165_strides_0 = const()[name = string("input_165_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_165_dilations_0 = const()[name = string("input_165_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39120128)))];
            tensor<fp16, [1024]> const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39138624)))];
            tensor<fp16, [1, 1024, 188]> input_167_cast_fp16 = conv(bias = const_48_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = const_47_to_fp16, x = input_163_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_23_cast_fp16 = silu(x = input_167_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<int32, [3]> var_3241_perm_0 = const()[name = string("op_3241_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_3_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39140736))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39402944))))[name = string("e_layers_3_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_3241_cast_fp16 = transpose(perm = var_3241_perm_0, x = x_23_cast_fp16)[name = string("transpose_121")];
            tensor<fp16, [1, 188, 1024]> var_3243_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_conv_pointwise_conv2_weight_to_fp16_lut, x = var_3241_cast_fp16)[name = string("op_3243_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_169_cast_fp16 = add(x = input_159_cast_fp16, y = var_3243_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<int32, [1]> input_171_axes_0 = const()[name = string("input_171_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39468544)))];
            tensor<fp16, [1024]> e_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39470656)))];
            fp16 var_3251_to_fp16 = const()[name = string("op_3251_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_171_cast_fp16 = layer_norm(axes = input_171_axes_0, beta = e_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_3251_to_fp16, gamma = e_layers_3_norm_feed_forward2_weight_to_fp16, x = input_169_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_3_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39472768))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40521408))))[name = string("e_layers_3_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_3_feed_forward2_linear1_weight_to_fp16_lut, x = input_171_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_175_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_175_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_3_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40783616))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41832256))))[name = string("e_layers_3_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_3_feed_forward2_linear2_weight_to_fp16_lut, x = input_175_cast_fp16)[name = string("linear_36_cast_fp16")];
            fp16 var_3268_to_fp16 = const()[name = string("op_3268_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3269_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_3268_to_fp16)[name = string("op_3269_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_179_cast_fp16 = add(x = input_169_cast_fp16, y = var_3269_cast_fp16)[name = string("input_179_cast_fp16")];
            tensor<int32, [1]> input_181_axes_0 = const()[name = string("input_181_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_3_norm_out_weight_to_fp16 = const()[name = string("e_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42094464)))];
            tensor<fp16, [1024]> e_layers_3_norm_out_bias_to_fp16 = const()[name = string("e_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42096576)))];
            fp16 var_3273_to_fp16 = const()[name = string("op_3273_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_181_cast_fp16 = layer_norm(axes = input_181_axes_0, beta = e_layers_3_norm_out_bias_to_fp16, epsilon = var_3273_to_fp16, gamma = e_layers_3_norm_out_weight_to_fp16, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<int32, [1]> input_183_axes_0 = const()[name = string("input_183_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42098688)))];
            tensor<fp16, [1024]> e_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42100800)))];
            fp16 var_3280_to_fp16 = const()[name = string("op_3280_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_183_cast_fp16 = layer_norm(axes = input_183_axes_0, beta = e_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_3280_to_fp16, gamma = e_layers_4_norm_feed_forward1_weight_to_fp16, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_4_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42102912))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43151552))))[name = string("e_layers_4_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_4_feed_forward1_linear1_weight_to_fp16_lut, x = input_183_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_187_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_187_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_4_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43413760))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44462400))))[name = string("e_layers_4_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_feed_forward1_linear2_weight_to_fp16_lut, x = input_187_cast_fp16)[name = string("linear_38_cast_fp16")];
            fp16 var_3297_to_fp16 = const()[name = string("op_3297_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3298_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_3297_to_fp16)[name = string("op_3298_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_191_cast_fp16 = add(x = input_181_cast_fp16, y = var_3298_cast_fp16)[name = string("input_191_cast_fp16")];
            tensor<int32, [1]> input_193_axes_0 = const()[name = string("input_193_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44724608)))];
            tensor<fp16, [1024]> e_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44726720)))];
            fp16 var_3302_to_fp16 = const()[name = string("op_3302_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_193_cast_fp16 = layer_norm(axes = input_193_axes_0, beta = e_layers_4_norm_self_att_bias_to_fp16, epsilon = var_3302_to_fp16, gamma = e_layers_4_norm_self_att_weight_to_fp16, x = input_191_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_4_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44728832))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44991040))))[name = string("e_layers_4_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_self_attn_q_proj_weight_to_fp16_lut, x = input_193_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> var_3315 = const()[name = string("op_3315"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3316_cast_fp16 = reshape(shape = var_3315, x = linear_39_cast_fp16)[name = string("op_3316_cast_fp16")];
            tensor<int32, [4]> q_9_perm_0 = const()[name = string("q_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_4_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45056640))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45318848))))[name = string("e_layers_4_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_self_attn_k_proj_weight_to_fp16_lut, x = input_193_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<int32, [4]> var_3327 = const()[name = string("op_3327"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3328_cast_fp16 = reshape(shape = var_3327, x = linear_40_cast_fp16)[name = string("op_3328_cast_fp16")];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_4_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45384448))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45646656))))[name = string("e_layers_4_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_self_attn_v_proj_weight_to_fp16_lut, x = input_193_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> var_3339 = const()[name = string("op_3339"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3340_cast_fp16 = reshape(shape = var_3339, x = linear_41_cast_fp16)[name = string("op_3340_cast_fp16")];
            tensor<int32, [4]> v_9_perm_0 = const()[name = string("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_3364_to_fp16 = const()[name = string("op_3364_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45712256)))];
            tensor<fp16, [1, 8, 188, 128]> q_9_cast_fp16 = transpose(perm = q_9_perm_0, x = var_3316_cast_fp16)[name = string("transpose_120")];
            tensor<fp16, [1, 8, 188, 128]> var_3366_cast_fp16 = add(x = q_9_cast_fp16, y = var_3364_to_fp16)[name = string("op_3366_cast_fp16")];
            bool input_195_transpose_x_0 = const()[name = string("input_195_transpose_x_0"), val = bool(false)];
            bool input_195_transpose_y_0 = const()[name = string("input_195_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_9_to_fp16 = const()[name = string("p_9_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45714368)))];
            tensor<fp16, [1, 8, 188, 375]> input_195_cast_fp16 = matmul(transpose_x = input_195_transpose_x_0, transpose_y = input_195_transpose_y_0, x = var_3366_cast_fp16, y = p_9_to_fp16)[name = string("input_195_cast_fp16")];
            tensor<int32, [8]> var_3373_pad_0 = const()[name = string("op_3373_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3373_mode_0 = const()[name = string("op_3373_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3373_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = var_3373_mode_0, pad = var_3373_pad_0, x = input_195_cast_fp16)[name = string("op_3373_cast_fp16")];
            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_3379_cast_fp16 = reshape(shape = var_3378, x = var_3373_cast_fp16)[name = string("op_3379_cast_fp16")];
            tensor<int32, [4]> var_3394_begin_0 = const()[name = string("op_3394_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3394_end_0 = const()[name = string("op_3394_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3394_end_mask_0 = const()[name = string("op_3394_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3394_cast_fp16 = slice_by_index(begin = var_3394_begin_0, end = var_3394_end_0, end_mask = var_3394_end_mask_0, x = var_3379_cast_fp16)[name = string("op_3394_cast_fp16")];
            tensor<int32, [4]> var_3399 = const()[name = string("op_3399"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3400_cast_fp16 = reshape(shape = var_3399, x = var_3394_cast_fp16)[name = string("op_3400_cast_fp16")];
            tensor<int32, [4]> bd_9_begin_0 = const()[name = string("bd_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_9_end_0 = const()[name = string("bd_9_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_9_end_mask_0 = const()[name = string("bd_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_9_cast_fp16 = slice_by_index(begin = bd_9_begin_0, end = bd_9_end_0, end_mask = bd_9_end_mask_0, x = var_3400_cast_fp16)[name = string("bd_9_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_3411_to_fp16 = const()[name = string("op_3411_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46482432)))];
            tensor<fp16, [1, 8, 188, 128]> var_3413_cast_fp16 = add(x = q_9_cast_fp16, y = var_3411_to_fp16)[name = string("op_3413_cast_fp16")];
            bool ac_9_transpose_x_0 = const()[name = string("ac_9_transpose_x_0"), val = bool(false)];
            bool ac_9_transpose_y_0 = const()[name = string("ac_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_52_cast_fp16 = transpose(perm = transpose_52_perm_0, x = var_3328_cast_fp16)[name = string("transpose_119")];
            tensor<fp16, [1, 8, 188, 188]> ac_9_cast_fp16 = matmul(transpose_x = ac_9_transpose_x_0, transpose_y = ac_9_transpose_y_0, x = var_3413_cast_fp16, y = transpose_52_cast_fp16)[name = string("ac_9_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3419_cast_fp16 = add(x = ac_9_cast_fp16, y = bd_9_cast_fp16)[name = string("op_3419_cast_fp16")];
            fp16 var_3420_to_fp16 = const()[name = string("op_3420_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3421_cast_fp16 = mul(x = var_3419_cast_fp16, y = var_3420_to_fp16)[name = string("op_3421_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3423_cast_fp16 = add(x = var_3421_cast_fp16, y = bias_15_cast_fp16)[name = string("op_3423_cast_fp16")];
            int32 var_3424 = const()[name = string("op_3424"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_9_cast_fp16 = softmax(axis = var_3424, x = var_3423_cast_fp16)[name = string("w_9_cast_fp16")];
            bool var_3427_transpose_x_0 = const()[name = string("op_3427_transpose_x_0"), val = bool(false)];
            bool var_3427_transpose_y_0 = const()[name = string("op_3427_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_3340_cast_fp16)[name = string("transpose_118")];
            tensor<fp16, [1, 8, 188, 128]> var_3427_cast_fp16 = matmul(transpose_x = var_3427_transpose_x_0, transpose_y = var_3427_transpose_y_0, x = w_9_cast_fp16, y = v_9_cast_fp16)[name = string("op_3427_cast_fp16")];
            tensor<int32, [4]> var_3430_perm_0 = const()[name = string("op_3430_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3434 = const()[name = string("op_3434"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3430_cast_fp16 = transpose(perm = var_3430_perm_0, x = var_3427_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [1, 188, 1024]> input_197_cast_fp16 = reshape(shape = var_3434, x = var_3430_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_4_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46484544))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46746752))))[name = string("e_layers_4_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_self_attn_o_proj_weight_to_fp16_lut, x = input_197_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_199_cast_fp16 = add(x = input_191_cast_fp16, y = linear_43_cast_fp16)[name = string("input_199_cast_fp16")];
            tensor<int32, [1]> x_25_axes_0 = const()[name = string("x_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_4_norm_conv_weight_to_fp16 = const()[name = string("e_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46812352)))];
            tensor<fp16, [1024]> e_layers_4_norm_conv_bias_to_fp16 = const()[name = string("e_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46814464)))];
            fp16 var_3442_to_fp16 = const()[name = string("op_3442_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_25_cast_fp16 = layer_norm(axes = x_25_axes_0, beta = e_layers_4_norm_conv_bias_to_fp16, epsilon = var_3442_to_fp16, gamma = e_layers_4_norm_conv_weight_to_fp16, x = input_199_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_4_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46816576))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47340928))))[name = string("e_layers_4_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_3456_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_4_conv_pointwise_conv1_weight_to_fp16_lut, x = x_25_cast_fp16)[name = string("op_3456_cast_fp16")];
            tensor<int32, [3]> input_201_perm_0 = const()[name = string("input_201_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_3459_split_num_splits_0 = const()[name = string("op_3459_split_num_splits_0"), val = int32(2)];
            int32 var_3459_split_axis_0 = const()[name = string("op_3459_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_201_cast_fp16 = transpose(perm = input_201_perm_0, x = var_3456_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [1, 1024, 188]> var_3459_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_3459_split_cast_fp16_1 = split(axis = var_3459_split_axis_0, num_splits = var_3459_split_num_splits_0, x = input_201_cast_fp16)[name = string("op_3459_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3459_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3459_split_cast_fp16_1)[name = string("op_3459_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3459_cast_fp16 = mul(x = var_3459_split_cast_fp16_0, y = var_3459_split_1_sigmoid_cast_fp16)[name = string("op_3459_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_203_cast_fp16 = mul(x = var_3459_cast_fp16, y = var_2228_cast_fp16)[name = string("input_203_cast_fp16")];
            string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47472064)))];
            tensor<fp16, [1024]> const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47490560)))];
            tensor<fp16, [1, 1024, 188]> input_207_cast_fp16 = conv(bias = const_50_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_49_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_29_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<int32, [3]> var_3499_perm_0 = const()[name = string("op_3499_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_4_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47492672))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47754880))))[name = string("e_layers_4_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_3499_cast_fp16 = transpose(perm = var_3499_perm_0, x = x_29_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [1, 188, 1024]> var_3501_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_conv_pointwise_conv2_weight_to_fp16_lut, x = var_3499_cast_fp16)[name = string("op_3501_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_209_cast_fp16 = add(x = input_199_cast_fp16, y = var_3501_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<int32, [1]> input_211_axes_0 = const()[name = string("input_211_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47820480)))];
            tensor<fp16, [1024]> e_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47822592)))];
            fp16 var_3509_to_fp16 = const()[name = string("op_3509_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_211_cast_fp16 = layer_norm(axes = input_211_axes_0, beta = e_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_3509_to_fp16, gamma = e_layers_4_norm_feed_forward2_weight_to_fp16, x = input_209_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_4_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47824704))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48873344))))[name = string("e_layers_4_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_4_feed_forward2_linear1_weight_to_fp16_lut, x = input_211_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_215_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_215_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_4_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49135552))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50184192))))[name = string("e_layers_4_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_4_feed_forward2_linear2_weight_to_fp16_lut, x = input_215_cast_fp16)[name = string("linear_45_cast_fp16")];
            fp16 var_3526_to_fp16 = const()[name = string("op_3526_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3527_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_3526_to_fp16)[name = string("op_3527_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_219_cast_fp16 = add(x = input_209_cast_fp16, y = var_3527_cast_fp16)[name = string("input_219_cast_fp16")];
            tensor<int32, [1]> input_221_axes_0 = const()[name = string("input_221_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_4_norm_out_weight_to_fp16 = const()[name = string("e_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50446400)))];
            tensor<fp16, [1024]> e_layers_4_norm_out_bias_to_fp16 = const()[name = string("e_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50448512)))];
            fp16 var_3531_to_fp16 = const()[name = string("op_3531_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = e_layers_4_norm_out_bias_to_fp16, epsilon = var_3531_to_fp16, gamma = e_layers_4_norm_out_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")];
            tensor<int32, [1]> input_223_axes_0 = const()[name = string("input_223_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50450624)))];
            tensor<fp16, [1024]> e_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50452736)))];
            fp16 var_3538_to_fp16 = const()[name = string("op_3538_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_223_cast_fp16 = layer_norm(axes = input_223_axes_0, beta = e_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_3538_to_fp16, gamma = e_layers_5_norm_feed_forward1_weight_to_fp16, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_5_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50454848))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51503488))))[name = string("e_layers_5_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_5_feed_forward1_linear1_weight_to_fp16_lut, x = input_223_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_227_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_5_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51765696))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52814336))))[name = string("e_layers_5_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_feed_forward1_linear2_weight_to_fp16_lut, x = input_227_cast_fp16)[name = string("linear_47_cast_fp16")];
            fp16 var_3555_to_fp16 = const()[name = string("op_3555_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3556_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_3555_to_fp16)[name = string("op_3556_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_231_cast_fp16 = add(x = input_221_cast_fp16, y = var_3556_cast_fp16)[name = string("input_231_cast_fp16")];
            tensor<int32, [1]> input_233_axes_0 = const()[name = string("input_233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53076544)))];
            tensor<fp16, [1024]> e_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53078656)))];
            fp16 var_3560_to_fp16 = const()[name = string("op_3560_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_233_cast_fp16 = layer_norm(axes = input_233_axes_0, beta = e_layers_5_norm_self_att_bias_to_fp16, epsilon = var_3560_to_fp16, gamma = e_layers_5_norm_self_att_weight_to_fp16, x = input_231_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_5_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53080768))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53342976))))[name = string("e_layers_5_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_self_attn_q_proj_weight_to_fp16_lut, x = input_233_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<int32, [4]> var_3573 = const()[name = string("op_3573"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3574_cast_fp16 = reshape(shape = var_3573, x = linear_48_cast_fp16)[name = string("op_3574_cast_fp16")];
            tensor<int32, [4]> q_11_perm_0 = const()[name = string("q_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_5_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53408576))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53670784))))[name = string("e_layers_5_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_self_attn_k_proj_weight_to_fp16_lut, x = input_233_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<int32, [4]> var_3585 = const()[name = string("op_3585"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3586_cast_fp16 = reshape(shape = var_3585, x = linear_49_cast_fp16)[name = string("op_3586_cast_fp16")];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_5_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53736384))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53998592))))[name = string("e_layers_5_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_self_attn_v_proj_weight_to_fp16_lut, x = input_233_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<int32, [4]> var_3597 = const()[name = string("op_3597"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3598_cast_fp16 = reshape(shape = var_3597, x = linear_50_cast_fp16)[name = string("op_3598_cast_fp16")];
            tensor<int32, [4]> v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_3622_to_fp16 = const()[name = string("op_3622_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54064192)))];
            tensor<fp16, [1, 8, 188, 128]> q_11_cast_fp16 = transpose(perm = q_11_perm_0, x = var_3574_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [1, 8, 188, 128]> var_3624_cast_fp16 = add(x = q_11_cast_fp16, y = var_3622_to_fp16)[name = string("op_3624_cast_fp16")];
            bool input_235_transpose_x_0 = const()[name = string("input_235_transpose_x_0"), val = bool(false)];
            bool input_235_transpose_y_0 = const()[name = string("input_235_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_11_to_fp16 = const()[name = string("p_11_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54066304)))];
            tensor<fp16, [1, 8, 188, 375]> input_235_cast_fp16 = matmul(transpose_x = input_235_transpose_x_0, transpose_y = input_235_transpose_y_0, x = var_3624_cast_fp16, y = p_11_to_fp16)[name = string("input_235_cast_fp16")];
            tensor<int32, [8]> var_3631_pad_0 = const()[name = string("op_3631_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3631_mode_0 = const()[name = string("op_3631_mode_0"), val = string("constant")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3631_cast_fp16 = pad(constant_val = const_22_to_fp16, mode = var_3631_mode_0, pad = var_3631_pad_0, x = input_235_cast_fp16)[name = string("op_3631_cast_fp16")];
            tensor<int32, [4]> var_3636 = const()[name = string("op_3636"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_3637_cast_fp16 = reshape(shape = var_3636, x = var_3631_cast_fp16)[name = string("op_3637_cast_fp16")];
            tensor<int32, [4]> var_3652_begin_0 = const()[name = string("op_3652_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3652_end_0 = const()[name = string("op_3652_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3652_end_mask_0 = const()[name = string("op_3652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3652_cast_fp16 = slice_by_index(begin = var_3652_begin_0, end = var_3652_end_0, end_mask = var_3652_end_mask_0, x = var_3637_cast_fp16)[name = string("op_3652_cast_fp16")];
            tensor<int32, [4]> var_3657 = const()[name = string("op_3657"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3658_cast_fp16 = reshape(shape = var_3657, x = var_3652_cast_fp16)[name = string("op_3658_cast_fp16")];
            tensor<int32, [4]> bd_11_begin_0 = const()[name = string("bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_11_end_0 = const()[name = string("bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_11_end_mask_0 = const()[name = string("bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_11_cast_fp16 = slice_by_index(begin = bd_11_begin_0, end = bd_11_end_0, end_mask = bd_11_end_mask_0, x = var_3658_cast_fp16)[name = string("bd_11_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_3669_to_fp16 = const()[name = string("op_3669_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54834368)))];
            tensor<fp16, [1, 8, 188, 128]> var_3671_cast_fp16 = add(x = q_11_cast_fp16, y = var_3669_to_fp16)[name = string("op_3671_cast_fp16")];
            bool ac_11_transpose_x_0 = const()[name = string("ac_11_transpose_x_0"), val = bool(false)];
            bool ac_11_transpose_y_0 = const()[name = string("ac_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_53_cast_fp16 = transpose(perm = transpose_53_perm_0, x = var_3586_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [1, 8, 188, 188]> ac_11_cast_fp16 = matmul(transpose_x = ac_11_transpose_x_0, transpose_y = ac_11_transpose_y_0, x = var_3671_cast_fp16, y = transpose_53_cast_fp16)[name = string("ac_11_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3677_cast_fp16 = add(x = ac_11_cast_fp16, y = bd_11_cast_fp16)[name = string("op_3677_cast_fp16")];
            fp16 var_3678_to_fp16 = const()[name = string("op_3678_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3679_cast_fp16 = mul(x = var_3677_cast_fp16, y = var_3678_to_fp16)[name = string("op_3679_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3681_cast_fp16 = add(x = var_3679_cast_fp16, y = bias_15_cast_fp16)[name = string("op_3681_cast_fp16")];
            int32 var_3682 = const()[name = string("op_3682"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_11_cast_fp16 = softmax(axis = var_3682, x = var_3681_cast_fp16)[name = string("w_11_cast_fp16")];
            bool var_3685_transpose_x_0 = const()[name = string("op_3685_transpose_x_0"), val = bool(false)];
            bool var_3685_transpose_y_0 = const()[name = string("op_3685_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_3598_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [1, 8, 188, 128]> var_3685_cast_fp16 = matmul(transpose_x = var_3685_transpose_x_0, transpose_y = var_3685_transpose_y_0, x = w_11_cast_fp16, y = v_11_cast_fp16)[name = string("op_3685_cast_fp16")];
            tensor<int32, [4]> var_3688_perm_0 = const()[name = string("op_3688_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3692 = const()[name = string("op_3692"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3688_cast_fp16 = transpose(perm = var_3688_perm_0, x = var_3685_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [1, 188, 1024]> input_237_cast_fp16 = reshape(shape = var_3692, x = var_3688_cast_fp16)[name = string("input_237_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_5_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54836480))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55098688))))[name = string("e_layers_5_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_self_attn_o_proj_weight_to_fp16_lut, x = input_237_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_239_cast_fp16 = add(x = input_231_cast_fp16, y = linear_52_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<int32, [1]> x_31_axes_0 = const()[name = string("x_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_5_norm_conv_weight_to_fp16 = const()[name = string("e_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55164288)))];
            tensor<fp16, [1024]> e_layers_5_norm_conv_bias_to_fp16 = const()[name = string("e_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55166400)))];
            fp16 var_3700_to_fp16 = const()[name = string("op_3700_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_31_cast_fp16 = layer_norm(axes = x_31_axes_0, beta = e_layers_5_norm_conv_bias_to_fp16, epsilon = var_3700_to_fp16, gamma = e_layers_5_norm_conv_weight_to_fp16, x = input_239_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_5_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55168512))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55692864))))[name = string("e_layers_5_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_3714_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_5_conv_pointwise_conv1_weight_to_fp16_lut, x = x_31_cast_fp16)[name = string("op_3714_cast_fp16")];
            tensor<int32, [3]> input_241_perm_0 = const()[name = string("input_241_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_3717_split_num_splits_0 = const()[name = string("op_3717_split_num_splits_0"), val = int32(2)];
            int32 var_3717_split_axis_0 = const()[name = string("op_3717_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_241_cast_fp16 = transpose(perm = input_241_perm_0, x = var_3714_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [1, 1024, 188]> var_3717_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_3717_split_cast_fp16_1 = split(axis = var_3717_split_axis_0, num_splits = var_3717_split_num_splits_0, x = input_241_cast_fp16)[name = string("op_3717_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3717_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3717_split_cast_fp16_1)[name = string("op_3717_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3717_cast_fp16 = mul(x = var_3717_split_cast_fp16_0, y = var_3717_split_1_sigmoid_cast_fp16)[name = string("op_3717_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_243_cast_fp16 = mul(x = var_3717_cast_fp16, y = var_2228_cast_fp16)[name = string("input_243_cast_fp16")];
            string input_245_pad_type_0 = const()[name = string("input_245_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_245_groups_0 = const()[name = string("input_245_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_245_strides_0 = const()[name = string("input_245_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_245_dilations_0 = const()[name = string("input_245_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55824000)))];
            tensor<fp16, [1024]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55842496)))];
            tensor<fp16, [1, 1024, 188]> input_247_cast_fp16 = conv(bias = const_52_to_fp16, dilations = input_245_dilations_0, groups = input_245_groups_0, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = input_245_strides_0, weight = const_51_to_fp16, x = input_243_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_35_cast_fp16 = silu(x = input_247_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [3]> var_3757_perm_0 = const()[name = string("op_3757_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_5_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55844608))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56106816))))[name = string("e_layers_5_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_3757_cast_fp16 = transpose(perm = var_3757_perm_0, x = x_35_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [1, 188, 1024]> var_3759_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_conv_pointwise_conv2_weight_to_fp16_lut, x = var_3757_cast_fp16)[name = string("op_3759_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_249_cast_fp16 = add(x = input_239_cast_fp16, y = var_3759_cast_fp16)[name = string("input_249_cast_fp16")];
            tensor<int32, [1]> input_251_axes_0 = const()[name = string("input_251_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56172416)))];
            tensor<fp16, [1024]> e_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56174528)))];
            fp16 var_3767_to_fp16 = const()[name = string("op_3767_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_251_cast_fp16 = layer_norm(axes = input_251_axes_0, beta = e_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_3767_to_fp16, gamma = e_layers_5_norm_feed_forward2_weight_to_fp16, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_5_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56176640))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57225280))))[name = string("e_layers_5_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_5_feed_forward2_linear1_weight_to_fp16_lut, x = input_251_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_255_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_255_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_5_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57487488))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58536128))))[name = string("e_layers_5_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_5_feed_forward2_linear2_weight_to_fp16_lut, x = input_255_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 var_3784_to_fp16 = const()[name = string("op_3784_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3785_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_3784_to_fp16)[name = string("op_3785_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_259_cast_fp16 = add(x = input_249_cast_fp16, y = var_3785_cast_fp16)[name = string("input_259_cast_fp16")];
            tensor<int32, [1]> input_261_axes_0 = const()[name = string("input_261_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_5_norm_out_weight_to_fp16 = const()[name = string("e_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58798336)))];
            tensor<fp16, [1024]> e_layers_5_norm_out_bias_to_fp16 = const()[name = string("e_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58800448)))];
            fp16 var_3789_to_fp16 = const()[name = string("op_3789_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_261_cast_fp16 = layer_norm(axes = input_261_axes_0, beta = e_layers_5_norm_out_bias_to_fp16, epsilon = var_3789_to_fp16, gamma = e_layers_5_norm_out_weight_to_fp16, x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
            tensor<int32, [1]> input_263_axes_0 = const()[name = string("input_263_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58802560)))];
            tensor<fp16, [1024]> e_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58804672)))];
            fp16 var_3796_to_fp16 = const()[name = string("op_3796_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_263_cast_fp16 = layer_norm(axes = input_263_axes_0, beta = e_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_3796_to_fp16, gamma = e_layers_6_norm_feed_forward1_weight_to_fp16, x = input_261_cast_fp16)[name = string("input_263_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_6_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58806784))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59855424))))[name = string("e_layers_6_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_6_feed_forward1_linear1_weight_to_fp16_lut, x = input_263_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_267_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_267_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_6_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60117632))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61166272))))[name = string("e_layers_6_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_feed_forward1_linear2_weight_to_fp16_lut, x = input_267_cast_fp16)[name = string("linear_56_cast_fp16")];
            fp16 var_3813_to_fp16 = const()[name = string("op_3813_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3814_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_3813_to_fp16)[name = string("op_3814_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_271_cast_fp16 = add(x = input_261_cast_fp16, y = var_3814_cast_fp16)[name = string("input_271_cast_fp16")];
            tensor<int32, [1]> input_273_axes_0 = const()[name = string("input_273_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61428480)))];
            tensor<fp16, [1024]> e_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61430592)))];
            fp16 var_3818_to_fp16 = const()[name = string("op_3818_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = e_layers_6_norm_self_att_bias_to_fp16, epsilon = var_3818_to_fp16, gamma = e_layers_6_norm_self_att_weight_to_fp16, x = input_271_cast_fp16)[name = string("input_273_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_6_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61432704))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61694912))))[name = string("e_layers_6_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_self_attn_q_proj_weight_to_fp16_lut, x = input_273_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<int32, [4]> var_3831 = const()[name = string("op_3831"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3832_cast_fp16 = reshape(shape = var_3831, x = linear_57_cast_fp16)[name = string("op_3832_cast_fp16")];
            tensor<int32, [4]> q_13_perm_0 = const()[name = string("q_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_6_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61760512))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62022720))))[name = string("e_layers_6_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_self_attn_k_proj_weight_to_fp16_lut, x = input_273_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<int32, [4]> var_3843 = const()[name = string("op_3843"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3844_cast_fp16 = reshape(shape = var_3843, x = linear_58_cast_fp16)[name = string("op_3844_cast_fp16")];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_6_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62088320))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62350528))))[name = string("e_layers_6_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_self_attn_v_proj_weight_to_fp16_lut, x = input_273_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<int32, [4]> var_3855 = const()[name = string("op_3855"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_3856_cast_fp16 = reshape(shape = var_3855, x = linear_59_cast_fp16)[name = string("op_3856_cast_fp16")];
            tensor<int32, [4]> v_13_perm_0 = const()[name = string("v_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_3880_to_fp16 = const()[name = string("op_3880_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62416128)))];
            tensor<fp16, [1, 8, 188, 128]> q_13_cast_fp16 = transpose(perm = q_13_perm_0, x = var_3832_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [1, 8, 188, 128]> var_3882_cast_fp16 = add(x = q_13_cast_fp16, y = var_3880_to_fp16)[name = string("op_3882_cast_fp16")];
            bool input_275_transpose_x_0 = const()[name = string("input_275_transpose_x_0"), val = bool(false)];
            bool input_275_transpose_y_0 = const()[name = string("input_275_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_13_to_fp16 = const()[name = string("p_13_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62418240)))];
            tensor<fp16, [1, 8, 188, 375]> input_275_cast_fp16 = matmul(transpose_x = input_275_transpose_x_0, transpose_y = input_275_transpose_y_0, x = var_3882_cast_fp16, y = p_13_to_fp16)[name = string("input_275_cast_fp16")];
            tensor<int32, [8]> var_3889_pad_0 = const()[name = string("op_3889_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_3889_mode_0 = const()[name = string("op_3889_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_3889_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = var_3889_mode_0, pad = var_3889_pad_0, x = input_275_cast_fp16)[name = string("op_3889_cast_fp16")];
            tensor<int32, [4]> var_3894 = const()[name = string("op_3894"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_3895_cast_fp16 = reshape(shape = var_3894, x = var_3889_cast_fp16)[name = string("op_3895_cast_fp16")];
            tensor<int32, [4]> var_3910_begin_0 = const()[name = string("op_3910_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3910_end_0 = const()[name = string("op_3910_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3910_end_mask_0 = const()[name = string("op_3910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3910_cast_fp16 = slice_by_index(begin = var_3910_begin_0, end = var_3910_end_0, end_mask = var_3910_end_mask_0, x = var_3895_cast_fp16)[name = string("op_3910_cast_fp16")];
            tensor<int32, [4]> var_3915 = const()[name = string("op_3915"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_3916_cast_fp16 = reshape(shape = var_3915, x = var_3910_cast_fp16)[name = string("op_3916_cast_fp16")];
            tensor<int32, [4]> bd_13_begin_0 = const()[name = string("bd_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_13_end_0 = const()[name = string("bd_13_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_13_end_mask_0 = const()[name = string("bd_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_13_cast_fp16 = slice_by_index(begin = bd_13_begin_0, end = bd_13_end_0, end_mask = bd_13_end_mask_0, x = var_3916_cast_fp16)[name = string("bd_13_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_3927_to_fp16 = const()[name = string("op_3927_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63186304)))];
            tensor<fp16, [1, 8, 188, 128]> var_3929_cast_fp16 = add(x = q_13_cast_fp16, y = var_3927_to_fp16)[name = string("op_3929_cast_fp16")];
            bool ac_13_transpose_x_0 = const()[name = string("ac_13_transpose_x_0"), val = bool(false)];
            bool ac_13_transpose_y_0 = const()[name = string("ac_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_54_cast_fp16 = transpose(perm = transpose_54_perm_0, x = var_3844_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [1, 8, 188, 188]> ac_13_cast_fp16 = matmul(transpose_x = ac_13_transpose_x_0, transpose_y = ac_13_transpose_y_0, x = var_3929_cast_fp16, y = transpose_54_cast_fp16)[name = string("ac_13_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3935_cast_fp16 = add(x = ac_13_cast_fp16, y = bd_13_cast_fp16)[name = string("op_3935_cast_fp16")];
            fp16 var_3936_to_fp16 = const()[name = string("op_3936_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_3937_cast_fp16 = mul(x = var_3935_cast_fp16, y = var_3936_to_fp16)[name = string("op_3937_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3939_cast_fp16 = add(x = var_3937_cast_fp16, y = bias_15_cast_fp16)[name = string("op_3939_cast_fp16")];
            int32 var_3940 = const()[name = string("op_3940"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_13_cast_fp16 = softmax(axis = var_3940, x = var_3939_cast_fp16)[name = string("w_13_cast_fp16")];
            bool var_3943_transpose_x_0 = const()[name = string("op_3943_transpose_x_0"), val = bool(false)];
            bool var_3943_transpose_y_0 = const()[name = string("op_3943_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_3856_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [1, 8, 188, 128]> var_3943_cast_fp16 = matmul(transpose_x = var_3943_transpose_x_0, transpose_y = var_3943_transpose_y_0, x = w_13_cast_fp16, y = v_13_cast_fp16)[name = string("op_3943_cast_fp16")];
            tensor<int32, [4]> var_3946_perm_0 = const()[name = string("op_3946_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3950 = const()[name = string("op_3950"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3946_cast_fp16 = transpose(perm = var_3946_perm_0, x = var_3943_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [1, 188, 1024]> input_277_cast_fp16 = reshape(shape = var_3950, x = var_3946_cast_fp16)[name = string("input_277_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_6_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63188416))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63450624))))[name = string("e_layers_6_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_self_attn_o_proj_weight_to_fp16_lut, x = input_277_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_279_cast_fp16 = add(x = input_271_cast_fp16, y = linear_61_cast_fp16)[name = string("input_279_cast_fp16")];
            tensor<int32, [1]> x_37_axes_0 = const()[name = string("x_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_6_norm_conv_weight_to_fp16 = const()[name = string("e_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63516224)))];
            tensor<fp16, [1024]> e_layers_6_norm_conv_bias_to_fp16 = const()[name = string("e_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63518336)))];
            fp16 var_3958_to_fp16 = const()[name = string("op_3958_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_37_cast_fp16 = layer_norm(axes = x_37_axes_0, beta = e_layers_6_norm_conv_bias_to_fp16, epsilon = var_3958_to_fp16, gamma = e_layers_6_norm_conv_weight_to_fp16, x = input_279_cast_fp16)[name = string("x_37_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_6_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63520448))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64044800))))[name = string("e_layers_6_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_3972_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_6_conv_pointwise_conv1_weight_to_fp16_lut, x = x_37_cast_fp16)[name = string("op_3972_cast_fp16")];
            tensor<int32, [3]> input_281_perm_0 = const()[name = string("input_281_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_3975_split_num_splits_0 = const()[name = string("op_3975_split_num_splits_0"), val = int32(2)];
            int32 var_3975_split_axis_0 = const()[name = string("op_3975_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_281_cast_fp16 = transpose(perm = input_281_perm_0, x = var_3972_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [1, 1024, 188]> var_3975_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_3975_split_cast_fp16_1 = split(axis = var_3975_split_axis_0, num_splits = var_3975_split_num_splits_0, x = input_281_cast_fp16)[name = string("op_3975_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3975_split_1_sigmoid_cast_fp16 = sigmoid(x = var_3975_split_cast_fp16_1)[name = string("op_3975_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_3975_cast_fp16 = mul(x = var_3975_split_cast_fp16_0, y = var_3975_split_1_sigmoid_cast_fp16)[name = string("op_3975_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_283_cast_fp16 = mul(x = var_3975_cast_fp16, y = var_2228_cast_fp16)[name = string("input_283_cast_fp16")];
            string input_285_pad_type_0 = const()[name = string("input_285_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_285_pad_0 = const()[name = string("input_285_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_285_groups_0 = const()[name = string("input_285_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_285_strides_0 = const()[name = string("input_285_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_285_dilations_0 = const()[name = string("input_285_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64175936)))];
            tensor<fp16, [1024]> const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64194432)))];
            tensor<fp16, [1, 1024, 188]> input_287_cast_fp16 = conv(bias = const_54_to_fp16, dilations = input_285_dilations_0, groups = input_285_groups_0, pad = input_285_pad_0, pad_type = input_285_pad_type_0, strides = input_285_strides_0, weight = const_53_to_fp16, x = input_283_cast_fp16)[name = string("input_287_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_41_cast_fp16 = silu(x = input_287_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<int32, [3]> var_4015_perm_0 = const()[name = string("op_4015_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_6_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64196544))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64458752))))[name = string("e_layers_6_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_4015_cast_fp16 = transpose(perm = var_4015_perm_0, x = x_41_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [1, 188, 1024]> var_4017_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_conv_pointwise_conv2_weight_to_fp16_lut, x = var_4015_cast_fp16)[name = string("op_4017_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_289_cast_fp16 = add(x = input_279_cast_fp16, y = var_4017_cast_fp16)[name = string("input_289_cast_fp16")];
            tensor<int32, [1]> input_291_axes_0 = const()[name = string("input_291_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64524352)))];
            tensor<fp16, [1024]> e_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64526464)))];
            fp16 var_4025_to_fp16 = const()[name = string("op_4025_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_291_cast_fp16 = layer_norm(axes = input_291_axes_0, beta = e_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_4025_to_fp16, gamma = e_layers_6_norm_feed_forward2_weight_to_fp16, x = input_289_cast_fp16)[name = string("input_291_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_6_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64528576))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65577216))))[name = string("e_layers_6_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_6_feed_forward2_linear1_weight_to_fp16_lut, x = input_291_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_295_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_295_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_6_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65839424))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66888064))))[name = string("e_layers_6_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_6_feed_forward2_linear2_weight_to_fp16_lut, x = input_295_cast_fp16)[name = string("linear_63_cast_fp16")];
            fp16 var_4042_to_fp16 = const()[name = string("op_4042_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4043_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_4042_to_fp16)[name = string("op_4043_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_299_cast_fp16 = add(x = input_289_cast_fp16, y = var_4043_cast_fp16)[name = string("input_299_cast_fp16")];
            tensor<int32, [1]> input_301_axes_0 = const()[name = string("input_301_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_6_norm_out_weight_to_fp16 = const()[name = string("e_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67150272)))];
            tensor<fp16, [1024]> e_layers_6_norm_out_bias_to_fp16 = const()[name = string("e_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67152384)))];
            fp16 var_4047_to_fp16 = const()[name = string("op_4047_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_301_cast_fp16 = layer_norm(axes = input_301_axes_0, beta = e_layers_6_norm_out_bias_to_fp16, epsilon = var_4047_to_fp16, gamma = e_layers_6_norm_out_weight_to_fp16, x = input_299_cast_fp16)[name = string("input_301_cast_fp16")];
            tensor<int32, [1]> input_303_axes_0 = const()[name = string("input_303_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67154496)))];
            tensor<fp16, [1024]> e_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67156608)))];
            fp16 var_4054_to_fp16 = const()[name = string("op_4054_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_303_cast_fp16 = layer_norm(axes = input_303_axes_0, beta = e_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_4054_to_fp16, gamma = e_layers_7_norm_feed_forward1_weight_to_fp16, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_7_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67158720))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68207360))))[name = string("e_layers_7_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_7_feed_forward1_linear1_weight_to_fp16_lut, x = input_303_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_307_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_307_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_7_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68469568))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69518208))))[name = string("e_layers_7_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_feed_forward1_linear2_weight_to_fp16_lut, x = input_307_cast_fp16)[name = string("linear_65_cast_fp16")];
            fp16 var_4071_to_fp16 = const()[name = string("op_4071_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4072_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_4071_to_fp16)[name = string("op_4072_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_311_cast_fp16 = add(x = input_301_cast_fp16, y = var_4072_cast_fp16)[name = string("input_311_cast_fp16")];
            tensor<int32, [1]> input_313_axes_0 = const()[name = string("input_313_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69780416)))];
            tensor<fp16, [1024]> e_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69782528)))];
            fp16 var_4076_to_fp16 = const()[name = string("op_4076_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_313_cast_fp16 = layer_norm(axes = input_313_axes_0, beta = e_layers_7_norm_self_att_bias_to_fp16, epsilon = var_4076_to_fp16, gamma = e_layers_7_norm_self_att_weight_to_fp16, x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_7_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69784640))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70046848))))[name = string("e_layers_7_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_self_attn_q_proj_weight_to_fp16_lut, x = input_313_cast_fp16)[name = string("linear_66_cast_fp16")];
            tensor<int32, [4]> var_4089 = const()[name = string("op_4089"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4090_cast_fp16 = reshape(shape = var_4089, x = linear_66_cast_fp16)[name = string("op_4090_cast_fp16")];
            tensor<int32, [4]> q_15_perm_0 = const()[name = string("q_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_7_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70112448))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70374656))))[name = string("e_layers_7_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_self_attn_k_proj_weight_to_fp16_lut, x = input_313_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<int32, [4]> var_4101 = const()[name = string("op_4101"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4102_cast_fp16 = reshape(shape = var_4101, x = linear_67_cast_fp16)[name = string("op_4102_cast_fp16")];
            tensor<int32, [4]> transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_7_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70440256))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70702464))))[name = string("e_layers_7_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_self_attn_v_proj_weight_to_fp16_lut, x = input_313_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [4]> var_4113 = const()[name = string("op_4113"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4114_cast_fp16 = reshape(shape = var_4113, x = linear_68_cast_fp16)[name = string("op_4114_cast_fp16")];
            tensor<int32, [4]> v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_4138_to_fp16 = const()[name = string("op_4138_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70768064)))];
            tensor<fp16, [1, 8, 188, 128]> q_15_cast_fp16 = transpose(perm = q_15_perm_0, x = var_4090_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [1, 8, 188, 128]> var_4140_cast_fp16 = add(x = q_15_cast_fp16, y = var_4138_to_fp16)[name = string("op_4140_cast_fp16")];
            bool input_315_transpose_x_0 = const()[name = string("input_315_transpose_x_0"), val = bool(false)];
            bool input_315_transpose_y_0 = const()[name = string("input_315_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_15_to_fp16 = const()[name = string("p_15_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70770176)))];
            tensor<fp16, [1, 8, 188, 375]> input_315_cast_fp16 = matmul(transpose_x = input_315_transpose_x_0, transpose_y = input_315_transpose_y_0, x = var_4140_cast_fp16, y = p_15_to_fp16)[name = string("input_315_cast_fp16")];
            tensor<int32, [8]> var_4147_pad_0 = const()[name = string("op_4147_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4147_mode_0 = const()[name = string("op_4147_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4147_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_4147_mode_0, pad = var_4147_pad_0, x = input_315_cast_fp16)[name = string("op_4147_cast_fp16")];
            tensor<int32, [4]> var_4152 = const()[name = string("op_4152"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_4153_cast_fp16 = reshape(shape = var_4152, x = var_4147_cast_fp16)[name = string("op_4153_cast_fp16")];
            tensor<int32, [4]> var_4168_begin_0 = const()[name = string("op_4168_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4168_end_0 = const()[name = string("op_4168_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4168_end_mask_0 = const()[name = string("op_4168_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4168_cast_fp16 = slice_by_index(begin = var_4168_begin_0, end = var_4168_end_0, end_mask = var_4168_end_mask_0, x = var_4153_cast_fp16)[name = string("op_4168_cast_fp16")];
            tensor<int32, [4]> var_4173 = const()[name = string("op_4173"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4174_cast_fp16 = reshape(shape = var_4173, x = var_4168_cast_fp16)[name = string("op_4174_cast_fp16")];
            tensor<int32, [4]> bd_15_begin_0 = const()[name = string("bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_15_end_0 = const()[name = string("bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_15_end_mask_0 = const()[name = string("bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_15_cast_fp16 = slice_by_index(begin = bd_15_begin_0, end = bd_15_end_0, end_mask = bd_15_end_mask_0, x = var_4174_cast_fp16)[name = string("bd_15_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_4185_to_fp16 = const()[name = string("op_4185_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71538240)))];
            tensor<fp16, [1, 8, 188, 128]> var_4187_cast_fp16 = add(x = q_15_cast_fp16, y = var_4185_to_fp16)[name = string("op_4187_cast_fp16")];
            bool ac_15_transpose_x_0 = const()[name = string("ac_15_transpose_x_0"), val = bool(false)];
            bool ac_15_transpose_y_0 = const()[name = string("ac_15_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_55_cast_fp16 = transpose(perm = transpose_55_perm_0, x = var_4102_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [1, 8, 188, 188]> ac_15_cast_fp16 = matmul(transpose_x = ac_15_transpose_x_0, transpose_y = ac_15_transpose_y_0, x = var_4187_cast_fp16, y = transpose_55_cast_fp16)[name = string("ac_15_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4193_cast_fp16 = add(x = ac_15_cast_fp16, y = bd_15_cast_fp16)[name = string("op_4193_cast_fp16")];
            fp16 var_4194_to_fp16 = const()[name = string("op_4194_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4195_cast_fp16 = mul(x = var_4193_cast_fp16, y = var_4194_to_fp16)[name = string("op_4195_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4197_cast_fp16 = add(x = var_4195_cast_fp16, y = bias_15_cast_fp16)[name = string("op_4197_cast_fp16")];
            int32 var_4198 = const()[name = string("op_4198"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_15_cast_fp16 = softmax(axis = var_4198, x = var_4197_cast_fp16)[name = string("w_15_cast_fp16")];
            bool var_4201_transpose_x_0 = const()[name = string("op_4201_transpose_x_0"), val = bool(false)];
            bool var_4201_transpose_y_0 = const()[name = string("op_4201_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_4114_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [1, 8, 188, 128]> var_4201_cast_fp16 = matmul(transpose_x = var_4201_transpose_x_0, transpose_y = var_4201_transpose_y_0, x = w_15_cast_fp16, y = v_15_cast_fp16)[name = string("op_4201_cast_fp16")];
            tensor<int32, [4]> var_4204_perm_0 = const()[name = string("op_4204_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4208 = const()[name = string("op_4208"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4204_cast_fp16 = transpose(perm = var_4204_perm_0, x = var_4201_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [1, 188, 1024]> input_317_cast_fp16 = reshape(shape = var_4208, x = var_4204_cast_fp16)[name = string("input_317_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_7_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71540352))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71802560))))[name = string("e_layers_7_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_self_attn_o_proj_weight_to_fp16_lut, x = input_317_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_319_cast_fp16 = add(x = input_311_cast_fp16, y = linear_70_cast_fp16)[name = string("input_319_cast_fp16")];
            tensor<int32, [1]> x_43_axes_0 = const()[name = string("x_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_7_norm_conv_weight_to_fp16 = const()[name = string("e_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71868160)))];
            tensor<fp16, [1024]> e_layers_7_norm_conv_bias_to_fp16 = const()[name = string("e_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71870272)))];
            fp16 var_4216_to_fp16 = const()[name = string("op_4216_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_43_cast_fp16 = layer_norm(axes = x_43_axes_0, beta = e_layers_7_norm_conv_bias_to_fp16, epsilon = var_4216_to_fp16, gamma = e_layers_7_norm_conv_weight_to_fp16, x = input_319_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_7_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71872384))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72396736))))[name = string("e_layers_7_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_4230_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_7_conv_pointwise_conv1_weight_to_fp16_lut, x = x_43_cast_fp16)[name = string("op_4230_cast_fp16")];
            tensor<int32, [3]> input_321_perm_0 = const()[name = string("input_321_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_4233_split_num_splits_0 = const()[name = string("op_4233_split_num_splits_0"), val = int32(2)];
            int32 var_4233_split_axis_0 = const()[name = string("op_4233_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_321_cast_fp16 = transpose(perm = input_321_perm_0, x = var_4230_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [1, 1024, 188]> var_4233_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_4233_split_cast_fp16_1 = split(axis = var_4233_split_axis_0, num_splits = var_4233_split_num_splits_0, x = input_321_cast_fp16)[name = string("op_4233_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4233_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4233_split_cast_fp16_1)[name = string("op_4233_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4233_cast_fp16 = mul(x = var_4233_split_cast_fp16_0, y = var_4233_split_1_sigmoid_cast_fp16)[name = string("op_4233_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_323_cast_fp16 = mul(x = var_4233_cast_fp16, y = var_2228_cast_fp16)[name = string("input_323_cast_fp16")];
            string input_325_pad_type_0 = const()[name = string("input_325_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_325_pad_0 = const()[name = string("input_325_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_325_groups_0 = const()[name = string("input_325_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_325_strides_0 = const()[name = string("input_325_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_325_dilations_0 = const()[name = string("input_325_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72527872)))];
            tensor<fp16, [1024]> const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72546368)))];
            tensor<fp16, [1, 1024, 188]> input_327_cast_fp16 = conv(bias = const_56_to_fp16, dilations = input_325_dilations_0, groups = input_325_groups_0, pad = input_325_pad_0, pad_type = input_325_pad_type_0, strides = input_325_strides_0, weight = const_55_to_fp16, x = input_323_cast_fp16)[name = string("input_327_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_47_cast_fp16 = silu(x = input_327_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<int32, [3]> var_4273_perm_0 = const()[name = string("op_4273_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_7_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72548480))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72810688))))[name = string("e_layers_7_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_4273_cast_fp16 = transpose(perm = var_4273_perm_0, x = x_47_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [1, 188, 1024]> var_4275_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_conv_pointwise_conv2_weight_to_fp16_lut, x = var_4273_cast_fp16)[name = string("op_4275_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_329_cast_fp16 = add(x = input_319_cast_fp16, y = var_4275_cast_fp16)[name = string("input_329_cast_fp16")];
            tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72876288)))];
            tensor<fp16, [1024]> e_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72878400)))];
            fp16 var_4283_to_fp16 = const()[name = string("op_4283_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = e_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_4283_to_fp16, gamma = e_layers_7_norm_feed_forward2_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_7_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72880512))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73929152))))[name = string("e_layers_7_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_7_feed_forward2_linear1_weight_to_fp16_lut, x = input_331_cast_fp16)[name = string("linear_71_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_335_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_335_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_7_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74191360))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75240000))))[name = string("e_layers_7_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_7_feed_forward2_linear2_weight_to_fp16_lut, x = input_335_cast_fp16)[name = string("linear_72_cast_fp16")];
            fp16 var_4300_to_fp16 = const()[name = string("op_4300_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4301_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_4300_to_fp16)[name = string("op_4301_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_339_cast_fp16 = add(x = input_329_cast_fp16, y = var_4301_cast_fp16)[name = string("input_339_cast_fp16")];
            tensor<int32, [1]> input_341_axes_0 = const()[name = string("input_341_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_7_norm_out_weight_to_fp16 = const()[name = string("e_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75502208)))];
            tensor<fp16, [1024]> e_layers_7_norm_out_bias_to_fp16 = const()[name = string("e_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75504320)))];
            fp16 var_4305_to_fp16 = const()[name = string("op_4305_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_341_cast_fp16 = layer_norm(axes = input_341_axes_0, beta = e_layers_7_norm_out_bias_to_fp16, epsilon = var_4305_to_fp16, gamma = e_layers_7_norm_out_weight_to_fp16, x = input_339_cast_fp16)[name = string("input_341_cast_fp16")];
            tensor<int32, [1]> input_343_axes_0 = const()[name = string("input_343_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75506432)))];
            tensor<fp16, [1024]> e_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75508544)))];
            fp16 var_4312_to_fp16 = const()[name = string("op_4312_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_343_cast_fp16 = layer_norm(axes = input_343_axes_0, beta = e_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_4312_to_fp16, gamma = e_layers_8_norm_feed_forward1_weight_to_fp16, x = input_341_cast_fp16)[name = string("input_343_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_8_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75510656))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76559296))))[name = string("e_layers_8_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_8_feed_forward1_linear1_weight_to_fp16_lut, x = input_343_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_347_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_8_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76821504))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77870144))))[name = string("e_layers_8_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_feed_forward1_linear2_weight_to_fp16_lut, x = input_347_cast_fp16)[name = string("linear_74_cast_fp16")];
            fp16 var_4329_to_fp16 = const()[name = string("op_4329_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4330_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_4329_to_fp16)[name = string("op_4330_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_351_cast_fp16 = add(x = input_341_cast_fp16, y = var_4330_cast_fp16)[name = string("input_351_cast_fp16")];
            tensor<int32, [1]> input_353_axes_0 = const()[name = string("input_353_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78132352)))];
            tensor<fp16, [1024]> e_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78134464)))];
            fp16 var_4334_to_fp16 = const()[name = string("op_4334_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_353_cast_fp16 = layer_norm(axes = input_353_axes_0, beta = e_layers_8_norm_self_att_bias_to_fp16, epsilon = var_4334_to_fp16, gamma = e_layers_8_norm_self_att_weight_to_fp16, x = input_351_cast_fp16)[name = string("input_353_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_8_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78136576))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78398784))))[name = string("e_layers_8_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_self_attn_q_proj_weight_to_fp16_lut, x = input_353_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<int32, [4]> var_4347 = const()[name = string("op_4347"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4348_cast_fp16 = reshape(shape = var_4347, x = linear_75_cast_fp16)[name = string("op_4348_cast_fp16")];
            tensor<int32, [4]> q_17_perm_0 = const()[name = string("q_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_8_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78464384))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78726592))))[name = string("e_layers_8_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_self_attn_k_proj_weight_to_fp16_lut, x = input_353_cast_fp16)[name = string("linear_76_cast_fp16")];
            tensor<int32, [4]> var_4359 = const()[name = string("op_4359"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4360_cast_fp16 = reshape(shape = var_4359, x = linear_76_cast_fp16)[name = string("op_4360_cast_fp16")];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_8_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78792192))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79054400))))[name = string("e_layers_8_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_self_attn_v_proj_weight_to_fp16_lut, x = input_353_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<int32, [4]> var_4371 = const()[name = string("op_4371"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4372_cast_fp16 = reshape(shape = var_4371, x = linear_77_cast_fp16)[name = string("op_4372_cast_fp16")];
            tensor<int32, [4]> v_17_perm_0 = const()[name = string("v_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_4396_to_fp16 = const()[name = string("op_4396_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79120000)))];
            tensor<fp16, [1, 8, 188, 128]> q_17_cast_fp16 = transpose(perm = q_17_perm_0, x = var_4348_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [1, 8, 188, 128]> var_4398_cast_fp16 = add(x = q_17_cast_fp16, y = var_4396_to_fp16)[name = string("op_4398_cast_fp16")];
            bool input_355_transpose_x_0 = const()[name = string("input_355_transpose_x_0"), val = bool(false)];
            bool input_355_transpose_y_0 = const()[name = string("input_355_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_17_to_fp16 = const()[name = string("p_17_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79122112)))];
            tensor<fp16, [1, 8, 188, 375]> input_355_cast_fp16 = matmul(transpose_x = input_355_transpose_x_0, transpose_y = input_355_transpose_y_0, x = var_4398_cast_fp16, y = p_17_to_fp16)[name = string("input_355_cast_fp16")];
            tensor<int32, [8]> var_4405_pad_0 = const()[name = string("op_4405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4405_mode_0 = const()[name = string("op_4405_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4405_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_4405_mode_0, pad = var_4405_pad_0, x = input_355_cast_fp16)[name = string("op_4405_cast_fp16")];
            tensor<int32, [4]> var_4410 = const()[name = string("op_4410"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_4411_cast_fp16 = reshape(shape = var_4410, x = var_4405_cast_fp16)[name = string("op_4411_cast_fp16")];
            tensor<int32, [4]> var_4426_begin_0 = const()[name = string("op_4426_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4426_end_0 = const()[name = string("op_4426_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4426_end_mask_0 = const()[name = string("op_4426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4426_cast_fp16 = slice_by_index(begin = var_4426_begin_0, end = var_4426_end_0, end_mask = var_4426_end_mask_0, x = var_4411_cast_fp16)[name = string("op_4426_cast_fp16")];
            tensor<int32, [4]> var_4431 = const()[name = string("op_4431"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4432_cast_fp16 = reshape(shape = var_4431, x = var_4426_cast_fp16)[name = string("op_4432_cast_fp16")];
            tensor<int32, [4]> bd_17_begin_0 = const()[name = string("bd_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_17_end_0 = const()[name = string("bd_17_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_17_end_mask_0 = const()[name = string("bd_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_17_cast_fp16 = slice_by_index(begin = bd_17_begin_0, end = bd_17_end_0, end_mask = bd_17_end_mask_0, x = var_4432_cast_fp16)[name = string("bd_17_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_4443_to_fp16 = const()[name = string("op_4443_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79890176)))];
            tensor<fp16, [1, 8, 188, 128]> var_4445_cast_fp16 = add(x = q_17_cast_fp16, y = var_4443_to_fp16)[name = string("op_4445_cast_fp16")];
            bool ac_17_transpose_x_0 = const()[name = string("ac_17_transpose_x_0"), val = bool(false)];
            bool ac_17_transpose_y_0 = const()[name = string("ac_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_56_cast_fp16 = transpose(perm = transpose_56_perm_0, x = var_4360_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [1, 8, 188, 188]> ac_17_cast_fp16 = matmul(transpose_x = ac_17_transpose_x_0, transpose_y = ac_17_transpose_y_0, x = var_4445_cast_fp16, y = transpose_56_cast_fp16)[name = string("ac_17_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4451_cast_fp16 = add(x = ac_17_cast_fp16, y = bd_17_cast_fp16)[name = string("op_4451_cast_fp16")];
            fp16 var_4452_to_fp16 = const()[name = string("op_4452_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4453_cast_fp16 = mul(x = var_4451_cast_fp16, y = var_4452_to_fp16)[name = string("op_4453_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4455_cast_fp16 = add(x = var_4453_cast_fp16, y = bias_15_cast_fp16)[name = string("op_4455_cast_fp16")];
            int32 var_4456 = const()[name = string("op_4456"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_17_cast_fp16 = softmax(axis = var_4456, x = var_4455_cast_fp16)[name = string("w_17_cast_fp16")];
            bool var_4459_transpose_x_0 = const()[name = string("op_4459_transpose_x_0"), val = bool(false)];
            bool var_4459_transpose_y_0 = const()[name = string("op_4459_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_17_cast_fp16 = transpose(perm = v_17_perm_0, x = var_4372_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [1, 8, 188, 128]> var_4459_cast_fp16 = matmul(transpose_x = var_4459_transpose_x_0, transpose_y = var_4459_transpose_y_0, x = w_17_cast_fp16, y = v_17_cast_fp16)[name = string("op_4459_cast_fp16")];
            tensor<int32, [4]> var_4462_perm_0 = const()[name = string("op_4462_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4466 = const()[name = string("op_4466"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4462_cast_fp16 = transpose(perm = var_4462_perm_0, x = var_4459_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [1, 188, 1024]> input_357_cast_fp16 = reshape(shape = var_4466, x = var_4462_cast_fp16)[name = string("input_357_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_8_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79892288))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80154496))))[name = string("e_layers_8_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_self_attn_o_proj_weight_to_fp16_lut, x = input_357_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_359_cast_fp16 = add(x = input_351_cast_fp16, y = linear_79_cast_fp16)[name = string("input_359_cast_fp16")];
            tensor<int32, [1]> x_49_axes_0 = const()[name = string("x_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_8_norm_conv_weight_to_fp16 = const()[name = string("e_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80220096)))];
            tensor<fp16, [1024]> e_layers_8_norm_conv_bias_to_fp16 = const()[name = string("e_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80222208)))];
            fp16 var_4474_to_fp16 = const()[name = string("op_4474_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_49_cast_fp16 = layer_norm(axes = x_49_axes_0, beta = e_layers_8_norm_conv_bias_to_fp16, epsilon = var_4474_to_fp16, gamma = e_layers_8_norm_conv_weight_to_fp16, x = input_359_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_8_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80224320))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80748672))))[name = string("e_layers_8_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_4488_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_8_conv_pointwise_conv1_weight_to_fp16_lut, x = x_49_cast_fp16)[name = string("op_4488_cast_fp16")];
            tensor<int32, [3]> input_361_perm_0 = const()[name = string("input_361_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_4491_split_num_splits_0 = const()[name = string("op_4491_split_num_splits_0"), val = int32(2)];
            int32 var_4491_split_axis_0 = const()[name = string("op_4491_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_361_cast_fp16 = transpose(perm = input_361_perm_0, x = var_4488_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [1, 1024, 188]> var_4491_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_4491_split_cast_fp16_1 = split(axis = var_4491_split_axis_0, num_splits = var_4491_split_num_splits_0, x = input_361_cast_fp16)[name = string("op_4491_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4491_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4491_split_cast_fp16_1)[name = string("op_4491_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4491_cast_fp16 = mul(x = var_4491_split_cast_fp16_0, y = var_4491_split_1_sigmoid_cast_fp16)[name = string("op_4491_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_363_cast_fp16 = mul(x = var_4491_cast_fp16, y = var_2228_cast_fp16)[name = string("input_363_cast_fp16")];
            string input_365_pad_type_0 = const()[name = string("input_365_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_365_pad_0 = const()[name = string("input_365_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_365_groups_0 = const()[name = string("input_365_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_365_strides_0 = const()[name = string("input_365_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_365_dilations_0 = const()[name = string("input_365_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80879808)))];
            tensor<fp16, [1024]> const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80898304)))];
            tensor<fp16, [1, 1024, 188]> input_367_cast_fp16 = conv(bias = const_58_to_fp16, dilations = input_365_dilations_0, groups = input_365_groups_0, pad = input_365_pad_0, pad_type = input_365_pad_type_0, strides = input_365_strides_0, weight = const_57_to_fp16, x = input_363_cast_fp16)[name = string("input_367_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_53_cast_fp16 = silu(x = input_367_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<int32, [3]> var_4531_perm_0 = const()[name = string("op_4531_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_8_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80900416))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81162624))))[name = string("e_layers_8_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_4531_cast_fp16 = transpose(perm = var_4531_perm_0, x = x_53_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [1, 188, 1024]> var_4533_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_conv_pointwise_conv2_weight_to_fp16_lut, x = var_4531_cast_fp16)[name = string("op_4533_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_369_cast_fp16 = add(x = input_359_cast_fp16, y = var_4533_cast_fp16)[name = string("input_369_cast_fp16")];
            tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81228224)))];
            tensor<fp16, [1024]> e_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81230336)))];
            fp16 var_4541_to_fp16 = const()[name = string("op_4541_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = e_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_4541_to_fp16, gamma = e_layers_8_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_8_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81232448))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82281088))))[name = string("e_layers_8_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_8_feed_forward2_linear1_weight_to_fp16_lut, x = input_371_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_375_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_375_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_8_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82543296))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83591936))))[name = string("e_layers_8_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_8_feed_forward2_linear2_weight_to_fp16_lut, x = input_375_cast_fp16)[name = string("linear_81_cast_fp16")];
            fp16 var_4558_to_fp16 = const()[name = string("op_4558_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4559_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_4558_to_fp16)[name = string("op_4559_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_379_cast_fp16 = add(x = input_369_cast_fp16, y = var_4559_cast_fp16)[name = string("input_379_cast_fp16")];
            tensor<int32, [1]> input_381_axes_0 = const()[name = string("input_381_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_8_norm_out_weight_to_fp16 = const()[name = string("e_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83854144)))];
            tensor<fp16, [1024]> e_layers_8_norm_out_bias_to_fp16 = const()[name = string("e_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83856256)))];
            fp16 var_4563_to_fp16 = const()[name = string("op_4563_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_381_cast_fp16 = layer_norm(axes = input_381_axes_0, beta = e_layers_8_norm_out_bias_to_fp16, epsilon = var_4563_to_fp16, gamma = e_layers_8_norm_out_weight_to_fp16, x = input_379_cast_fp16)[name = string("input_381_cast_fp16")];
            tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83858368)))];
            tensor<fp16, [1024]> e_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83860480)))];
            fp16 var_4570_to_fp16 = const()[name = string("op_4570_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = e_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_4570_to_fp16, gamma = e_layers_9_norm_feed_forward1_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_9_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83862592))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84911232))))[name = string("e_layers_9_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_9_feed_forward1_linear1_weight_to_fp16_lut, x = input_383_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_387_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_387_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_9_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85173440))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86222080))))[name = string("e_layers_9_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_feed_forward1_linear2_weight_to_fp16_lut, x = input_387_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 var_4587_to_fp16 = const()[name = string("op_4587_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4588_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_4587_to_fp16)[name = string("op_4588_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_391_cast_fp16 = add(x = input_381_cast_fp16, y = var_4588_cast_fp16)[name = string("input_391_cast_fp16")];
            tensor<int32, [1]> input_393_axes_0 = const()[name = string("input_393_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86484288)))];
            tensor<fp16, [1024]> e_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86486400)))];
            fp16 var_4592_to_fp16 = const()[name = string("op_4592_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_393_cast_fp16 = layer_norm(axes = input_393_axes_0, beta = e_layers_9_norm_self_att_bias_to_fp16, epsilon = var_4592_to_fp16, gamma = e_layers_9_norm_self_att_weight_to_fp16, x = input_391_cast_fp16)[name = string("input_393_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_9_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86488512))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86750720))))[name = string("e_layers_9_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_self_attn_q_proj_weight_to_fp16_lut, x = input_393_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<int32, [4]> var_4605 = const()[name = string("op_4605"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4606_cast_fp16 = reshape(shape = var_4605, x = linear_84_cast_fp16)[name = string("op_4606_cast_fp16")];
            tensor<int32, [4]> q_19_perm_0 = const()[name = string("q_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_9_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86816320))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87078528))))[name = string("e_layers_9_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_self_attn_k_proj_weight_to_fp16_lut, x = input_393_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> var_4617 = const()[name = string("op_4617"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4618_cast_fp16 = reshape(shape = var_4617, x = linear_85_cast_fp16)[name = string("op_4618_cast_fp16")];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_9_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87144128))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87406336))))[name = string("e_layers_9_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_self_attn_v_proj_weight_to_fp16_lut, x = input_393_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> var_4629 = const()[name = string("op_4629"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4630_cast_fp16 = reshape(shape = var_4629, x = linear_86_cast_fp16)[name = string("op_4630_cast_fp16")];
            tensor<int32, [4]> v_19_perm_0 = const()[name = string("v_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_4654_to_fp16 = const()[name = string("op_4654_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87471936)))];
            tensor<fp16, [1, 8, 188, 128]> q_19_cast_fp16 = transpose(perm = q_19_perm_0, x = var_4606_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [1, 8, 188, 128]> var_4656_cast_fp16 = add(x = q_19_cast_fp16, y = var_4654_to_fp16)[name = string("op_4656_cast_fp16")];
            bool input_395_transpose_x_0 = const()[name = string("input_395_transpose_x_0"), val = bool(false)];
            bool input_395_transpose_y_0 = const()[name = string("input_395_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_19_to_fp16 = const()[name = string("p_19_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87474048)))];
            tensor<fp16, [1, 8, 188, 375]> input_395_cast_fp16 = matmul(transpose_x = input_395_transpose_x_0, transpose_y = input_395_transpose_y_0, x = var_4656_cast_fp16, y = p_19_to_fp16)[name = string("input_395_cast_fp16")];
            tensor<int32, [8]> var_4663_pad_0 = const()[name = string("op_4663_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4663_mode_0 = const()[name = string("op_4663_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4663_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_4663_mode_0, pad = var_4663_pad_0, x = input_395_cast_fp16)[name = string("op_4663_cast_fp16")];
            tensor<int32, [4]> var_4668 = const()[name = string("op_4668"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_4669_cast_fp16 = reshape(shape = var_4668, x = var_4663_cast_fp16)[name = string("op_4669_cast_fp16")];
            tensor<int32, [4]> var_4684_begin_0 = const()[name = string("op_4684_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4684_end_0 = const()[name = string("op_4684_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4684_end_mask_0 = const()[name = string("op_4684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4684_cast_fp16 = slice_by_index(begin = var_4684_begin_0, end = var_4684_end_0, end_mask = var_4684_end_mask_0, x = var_4669_cast_fp16)[name = string("op_4684_cast_fp16")];
            tensor<int32, [4]> var_4689 = const()[name = string("op_4689"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4690_cast_fp16 = reshape(shape = var_4689, x = var_4684_cast_fp16)[name = string("op_4690_cast_fp16")];
            tensor<int32, [4]> bd_19_begin_0 = const()[name = string("bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_19_end_0 = const()[name = string("bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_19_end_mask_0 = const()[name = string("bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_19_cast_fp16 = slice_by_index(begin = bd_19_begin_0, end = bd_19_end_0, end_mask = bd_19_end_mask_0, x = var_4690_cast_fp16)[name = string("bd_19_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_4701_to_fp16 = const()[name = string("op_4701_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88242112)))];
            tensor<fp16, [1, 8, 188, 128]> var_4703_cast_fp16 = add(x = q_19_cast_fp16, y = var_4701_to_fp16)[name = string("op_4703_cast_fp16")];
            bool ac_19_transpose_x_0 = const()[name = string("ac_19_transpose_x_0"), val = bool(false)];
            bool ac_19_transpose_y_0 = const()[name = string("ac_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_57_cast_fp16 = transpose(perm = transpose_57_perm_0, x = var_4618_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [1, 8, 188, 188]> ac_19_cast_fp16 = matmul(transpose_x = ac_19_transpose_x_0, transpose_y = ac_19_transpose_y_0, x = var_4703_cast_fp16, y = transpose_57_cast_fp16)[name = string("ac_19_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4709_cast_fp16 = add(x = ac_19_cast_fp16, y = bd_19_cast_fp16)[name = string("op_4709_cast_fp16")];
            fp16 var_4710_to_fp16 = const()[name = string("op_4710_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4711_cast_fp16 = mul(x = var_4709_cast_fp16, y = var_4710_to_fp16)[name = string("op_4711_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4713_cast_fp16 = add(x = var_4711_cast_fp16, y = bias_15_cast_fp16)[name = string("op_4713_cast_fp16")];
            int32 var_4714 = const()[name = string("op_4714"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_19_cast_fp16 = softmax(axis = var_4714, x = var_4713_cast_fp16)[name = string("w_19_cast_fp16")];
            bool var_4717_transpose_x_0 = const()[name = string("op_4717_transpose_x_0"), val = bool(false)];
            bool var_4717_transpose_y_0 = const()[name = string("op_4717_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_19_cast_fp16 = transpose(perm = v_19_perm_0, x = var_4630_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [1, 8, 188, 128]> var_4717_cast_fp16 = matmul(transpose_x = var_4717_transpose_x_0, transpose_y = var_4717_transpose_y_0, x = w_19_cast_fp16, y = v_19_cast_fp16)[name = string("op_4717_cast_fp16")];
            tensor<int32, [4]> var_4720_perm_0 = const()[name = string("op_4720_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4724 = const()[name = string("op_4724"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4720_cast_fp16 = transpose(perm = var_4720_perm_0, x = var_4717_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [1, 188, 1024]> input_397_cast_fp16 = reshape(shape = var_4724, x = var_4720_cast_fp16)[name = string("input_397_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_9_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88244224))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88506432))))[name = string("e_layers_9_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_self_attn_o_proj_weight_to_fp16_lut, x = input_397_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_399_cast_fp16 = add(x = input_391_cast_fp16, y = linear_88_cast_fp16)[name = string("input_399_cast_fp16")];
            tensor<int32, [1]> x_55_axes_0 = const()[name = string("x_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_9_norm_conv_weight_to_fp16 = const()[name = string("e_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88572032)))];
            tensor<fp16, [1024]> e_layers_9_norm_conv_bias_to_fp16 = const()[name = string("e_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88574144)))];
            fp16 var_4732_to_fp16 = const()[name = string("op_4732_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_55_cast_fp16 = layer_norm(axes = x_55_axes_0, beta = e_layers_9_norm_conv_bias_to_fp16, epsilon = var_4732_to_fp16, gamma = e_layers_9_norm_conv_weight_to_fp16, x = input_399_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_9_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88576256))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89100608))))[name = string("e_layers_9_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_4746_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_9_conv_pointwise_conv1_weight_to_fp16_lut, x = x_55_cast_fp16)[name = string("op_4746_cast_fp16")];
            tensor<int32, [3]> input_401_perm_0 = const()[name = string("input_401_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_4749_split_num_splits_0 = const()[name = string("op_4749_split_num_splits_0"), val = int32(2)];
            int32 var_4749_split_axis_0 = const()[name = string("op_4749_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_401_cast_fp16 = transpose(perm = input_401_perm_0, x = var_4746_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [1, 1024, 188]> var_4749_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_4749_split_cast_fp16_1 = split(axis = var_4749_split_axis_0, num_splits = var_4749_split_num_splits_0, x = input_401_cast_fp16)[name = string("op_4749_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4749_split_1_sigmoid_cast_fp16 = sigmoid(x = var_4749_split_cast_fp16_1)[name = string("op_4749_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_4749_cast_fp16 = mul(x = var_4749_split_cast_fp16_0, y = var_4749_split_1_sigmoid_cast_fp16)[name = string("op_4749_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_403_cast_fp16 = mul(x = var_4749_cast_fp16, y = var_2228_cast_fp16)[name = string("input_403_cast_fp16")];
            string input_405_pad_type_0 = const()[name = string("input_405_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_405_pad_0 = const()[name = string("input_405_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_405_groups_0 = const()[name = string("input_405_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_405_strides_0 = const()[name = string("input_405_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_405_dilations_0 = const()[name = string("input_405_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89231744)))];
            tensor<fp16, [1024]> const_60_to_fp16 = const()[name = string("const_60_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89250240)))];
            tensor<fp16, [1, 1024, 188]> input_407_cast_fp16 = conv(bias = const_60_to_fp16, dilations = input_405_dilations_0, groups = input_405_groups_0, pad = input_405_pad_0, pad_type = input_405_pad_type_0, strides = input_405_strides_0, weight = const_59_to_fp16, x = input_403_cast_fp16)[name = string("input_407_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_59_cast_fp16 = silu(x = input_407_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [3]> var_4789_perm_0 = const()[name = string("op_4789_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_9_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89252352))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89514560))))[name = string("e_layers_9_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_4789_cast_fp16 = transpose(perm = var_4789_perm_0, x = x_59_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [1, 188, 1024]> var_4791_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_conv_pointwise_conv2_weight_to_fp16_lut, x = var_4789_cast_fp16)[name = string("op_4791_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_409_cast_fp16 = add(x = input_399_cast_fp16, y = var_4791_cast_fp16)[name = string("input_409_cast_fp16")];
            tensor<int32, [1]> input_411_axes_0 = const()[name = string("input_411_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89580160)))];
            tensor<fp16, [1024]> e_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89582272)))];
            fp16 var_4799_to_fp16 = const()[name = string("op_4799_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_411_cast_fp16 = layer_norm(axes = input_411_axes_0, beta = e_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_4799_to_fp16, gamma = e_layers_9_norm_feed_forward2_weight_to_fp16, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_9_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89584384))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90633024))))[name = string("e_layers_9_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_9_feed_forward2_linear1_weight_to_fp16_lut, x = input_411_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_415_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_415_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_9_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90895232))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91943872))))[name = string("e_layers_9_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_9_feed_forward2_linear2_weight_to_fp16_lut, x = input_415_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 var_4816_to_fp16 = const()[name = string("op_4816_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4817_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_4816_to_fp16)[name = string("op_4817_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_419_cast_fp16 = add(x = input_409_cast_fp16, y = var_4817_cast_fp16)[name = string("input_419_cast_fp16")];
            tensor<int32, [1]> input_421_axes_0 = const()[name = string("input_421_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_9_norm_out_weight_to_fp16 = const()[name = string("e_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92206080)))];
            tensor<fp16, [1024]> e_layers_9_norm_out_bias_to_fp16 = const()[name = string("e_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92208192)))];
            fp16 var_4821_to_fp16 = const()[name = string("op_4821_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_421_cast_fp16 = layer_norm(axes = input_421_axes_0, beta = e_layers_9_norm_out_bias_to_fp16, epsilon = var_4821_to_fp16, gamma = e_layers_9_norm_out_weight_to_fp16, x = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
            tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92210304)))];
            tensor<fp16, [1024]> e_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92212416)))];
            fp16 var_4828_to_fp16 = const()[name = string("op_4828_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = e_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_4828_to_fp16, gamma = e_layers_10_norm_feed_forward1_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_10_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92214528))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93263168))))[name = string("e_layers_10_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_10_feed_forward1_linear1_weight_to_fp16_lut, x = input_423_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_427_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_427_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_10_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93525376))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94574016))))[name = string("e_layers_10_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_feed_forward1_linear2_weight_to_fp16_lut, x = input_427_cast_fp16)[name = string("linear_92_cast_fp16")];
            fp16 var_4845_to_fp16 = const()[name = string("op_4845_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4846_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_4845_to_fp16)[name = string("op_4846_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_431_cast_fp16 = add(x = input_421_cast_fp16, y = var_4846_cast_fp16)[name = string("input_431_cast_fp16")];
            tensor<int32, [1]> input_433_axes_0 = const()[name = string("input_433_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94836224)))];
            tensor<fp16, [1024]> e_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94838336)))];
            fp16 var_4850_to_fp16 = const()[name = string("op_4850_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_433_cast_fp16 = layer_norm(axes = input_433_axes_0, beta = e_layers_10_norm_self_att_bias_to_fp16, epsilon = var_4850_to_fp16, gamma = e_layers_10_norm_self_att_weight_to_fp16, x = input_431_cast_fp16)[name = string("input_433_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_10_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94840448))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95102656))))[name = string("e_layers_10_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_self_attn_q_proj_weight_to_fp16_lut, x = input_433_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> var_4863 = const()[name = string("op_4863"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4864_cast_fp16 = reshape(shape = var_4863, x = linear_93_cast_fp16)[name = string("op_4864_cast_fp16")];
            tensor<int32, [4]> q_21_perm_0 = const()[name = string("q_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_10_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95168256))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95430464))))[name = string("e_layers_10_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_self_attn_k_proj_weight_to_fp16_lut, x = input_433_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> var_4875 = const()[name = string("op_4875"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4876_cast_fp16 = reshape(shape = var_4875, x = linear_94_cast_fp16)[name = string("op_4876_cast_fp16")];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = string("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_10_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95496064))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95758272))))[name = string("e_layers_10_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_self_attn_v_proj_weight_to_fp16_lut, x = input_433_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> var_4887 = const()[name = string("op_4887"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_4888_cast_fp16 = reshape(shape = var_4887, x = linear_95_cast_fp16)[name = string("op_4888_cast_fp16")];
            tensor<int32, [4]> v_21_perm_0 = const()[name = string("v_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_4912_to_fp16 = const()[name = string("op_4912_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95823872)))];
            tensor<fp16, [1, 8, 188, 128]> q_21_cast_fp16 = transpose(perm = q_21_perm_0, x = var_4864_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [1, 8, 188, 128]> var_4914_cast_fp16 = add(x = q_21_cast_fp16, y = var_4912_to_fp16)[name = string("op_4914_cast_fp16")];
            bool input_435_transpose_x_0 = const()[name = string("input_435_transpose_x_0"), val = bool(false)];
            bool input_435_transpose_y_0 = const()[name = string("input_435_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_21_to_fp16 = const()[name = string("p_21_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95825984)))];
            tensor<fp16, [1, 8, 188, 375]> input_435_cast_fp16 = matmul(transpose_x = input_435_transpose_x_0, transpose_y = input_435_transpose_y_0, x = var_4914_cast_fp16, y = p_21_to_fp16)[name = string("input_435_cast_fp16")];
            tensor<int32, [8]> var_4921_pad_0 = const()[name = string("op_4921_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_4921_mode_0 = const()[name = string("op_4921_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_4921_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_4921_mode_0, pad = var_4921_pad_0, x = input_435_cast_fp16)[name = string("op_4921_cast_fp16")];
            tensor<int32, [4]> var_4926 = const()[name = string("op_4926"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_4927_cast_fp16 = reshape(shape = var_4926, x = var_4921_cast_fp16)[name = string("op_4927_cast_fp16")];
            tensor<int32, [4]> var_4942_begin_0 = const()[name = string("op_4942_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4942_end_0 = const()[name = string("op_4942_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4942_end_mask_0 = const()[name = string("op_4942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4942_cast_fp16 = slice_by_index(begin = var_4942_begin_0, end = var_4942_end_0, end_mask = var_4942_end_mask_0, x = var_4927_cast_fp16)[name = string("op_4942_cast_fp16")];
            tensor<int32, [4]> var_4947 = const()[name = string("op_4947"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_4948_cast_fp16 = reshape(shape = var_4947, x = var_4942_cast_fp16)[name = string("op_4948_cast_fp16")];
            tensor<int32, [4]> bd_21_begin_0 = const()[name = string("bd_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_21_end_0 = const()[name = string("bd_21_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_21_end_mask_0 = const()[name = string("bd_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_21_cast_fp16 = slice_by_index(begin = bd_21_begin_0, end = bd_21_end_0, end_mask = bd_21_end_mask_0, x = var_4948_cast_fp16)[name = string("bd_21_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_4959_to_fp16 = const()[name = string("op_4959_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96594048)))];
            tensor<fp16, [1, 8, 188, 128]> var_4961_cast_fp16 = add(x = q_21_cast_fp16, y = var_4959_to_fp16)[name = string("op_4961_cast_fp16")];
            bool ac_21_transpose_x_0 = const()[name = string("ac_21_transpose_x_0"), val = bool(false)];
            bool ac_21_transpose_y_0 = const()[name = string("ac_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_58_cast_fp16 = transpose(perm = transpose_58_perm_0, x = var_4876_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [1, 8, 188, 188]> ac_21_cast_fp16 = matmul(transpose_x = ac_21_transpose_x_0, transpose_y = ac_21_transpose_y_0, x = var_4961_cast_fp16, y = transpose_58_cast_fp16)[name = string("ac_21_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4967_cast_fp16 = add(x = ac_21_cast_fp16, y = bd_21_cast_fp16)[name = string("op_4967_cast_fp16")];
            fp16 var_4968_to_fp16 = const()[name = string("op_4968_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_4969_cast_fp16 = mul(x = var_4967_cast_fp16, y = var_4968_to_fp16)[name = string("op_4969_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4971_cast_fp16 = add(x = var_4969_cast_fp16, y = bias_15_cast_fp16)[name = string("op_4971_cast_fp16")];
            int32 var_4972 = const()[name = string("op_4972"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_21_cast_fp16 = softmax(axis = var_4972, x = var_4971_cast_fp16)[name = string("w_21_cast_fp16")];
            bool var_4975_transpose_x_0 = const()[name = string("op_4975_transpose_x_0"), val = bool(false)];
            bool var_4975_transpose_y_0 = const()[name = string("op_4975_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_21_cast_fp16 = transpose(perm = v_21_perm_0, x = var_4888_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [1, 8, 188, 128]> var_4975_cast_fp16 = matmul(transpose_x = var_4975_transpose_x_0, transpose_y = var_4975_transpose_y_0, x = w_21_cast_fp16, y = v_21_cast_fp16)[name = string("op_4975_cast_fp16")];
            tensor<int32, [4]> var_4978_perm_0 = const()[name = string("op_4978_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4982 = const()[name = string("op_4982"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4978_cast_fp16 = transpose(perm = var_4978_perm_0, x = var_4975_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [1, 188, 1024]> input_437_cast_fp16 = reshape(shape = var_4982, x = var_4978_cast_fp16)[name = string("input_437_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_10_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96596160))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96858368))))[name = string("e_layers_10_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_self_attn_o_proj_weight_to_fp16_lut, x = input_437_cast_fp16)[name = string("linear_97_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_439_cast_fp16 = add(x = input_431_cast_fp16, y = linear_97_cast_fp16)[name = string("input_439_cast_fp16")];
            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_10_norm_conv_weight_to_fp16 = const()[name = string("e_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96923968)))];
            tensor<fp16, [1024]> e_layers_10_norm_conv_bias_to_fp16 = const()[name = string("e_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96926080)))];
            fp16 var_4990_to_fp16 = const()[name = string("op_4990_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = e_layers_10_norm_conv_bias_to_fp16, epsilon = var_4990_to_fp16, gamma = e_layers_10_norm_conv_weight_to_fp16, x = input_439_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_10_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96928192))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97452544))))[name = string("e_layers_10_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_5004_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_10_conv_pointwise_conv1_weight_to_fp16_lut, x = x_61_cast_fp16)[name = string("op_5004_cast_fp16")];
            tensor<int32, [3]> input_441_perm_0 = const()[name = string("input_441_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_5007_split_num_splits_0 = const()[name = string("op_5007_split_num_splits_0"), val = int32(2)];
            int32 var_5007_split_axis_0 = const()[name = string("op_5007_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_441_cast_fp16 = transpose(perm = input_441_perm_0, x = var_5004_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [1, 1024, 188]> var_5007_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_5007_split_cast_fp16_1 = split(axis = var_5007_split_axis_0, num_splits = var_5007_split_num_splits_0, x = input_441_cast_fp16)[name = string("op_5007_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5007_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5007_split_cast_fp16_1)[name = string("op_5007_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5007_cast_fp16 = mul(x = var_5007_split_cast_fp16_0, y = var_5007_split_1_sigmoid_cast_fp16)[name = string("op_5007_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_443_cast_fp16 = mul(x = var_5007_cast_fp16, y = var_2228_cast_fp16)[name = string("input_443_cast_fp16")];
            string input_445_pad_type_0 = const()[name = string("input_445_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_445_pad_0 = const()[name = string("input_445_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_445_groups_0 = const()[name = string("input_445_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_445_strides_0 = const()[name = string("input_445_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_445_dilations_0 = const()[name = string("input_445_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97583680)))];
            tensor<fp16, [1024]> const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97602176)))];
            tensor<fp16, [1, 1024, 188]> input_447_cast_fp16 = conv(bias = const_62_to_fp16, dilations = input_445_dilations_0, groups = input_445_groups_0, pad = input_445_pad_0, pad_type = input_445_pad_type_0, strides = input_445_strides_0, weight = const_61_to_fp16, x = input_443_cast_fp16)[name = string("input_447_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_65_cast_fp16 = silu(x = input_447_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [3]> var_5047_perm_0 = const()[name = string("op_5047_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_10_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97604288))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97866496))))[name = string("e_layers_10_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_5047_cast_fp16 = transpose(perm = var_5047_perm_0, x = x_65_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [1, 188, 1024]> var_5049_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_conv_pointwise_conv2_weight_to_fp16_lut, x = var_5047_cast_fp16)[name = string("op_5049_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_449_cast_fp16 = add(x = input_439_cast_fp16, y = var_5049_cast_fp16)[name = string("input_449_cast_fp16")];
            tensor<int32, [1]> input_451_axes_0 = const()[name = string("input_451_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97932096)))];
            tensor<fp16, [1024]> e_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97934208)))];
            fp16 var_5057_to_fp16 = const()[name = string("op_5057_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_451_cast_fp16 = layer_norm(axes = input_451_axes_0, beta = e_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_5057_to_fp16, gamma = e_layers_10_norm_feed_forward2_weight_to_fp16, x = input_449_cast_fp16)[name = string("input_451_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_10_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97936320))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98984960))))[name = string("e_layers_10_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_10_feed_forward2_linear1_weight_to_fp16_lut, x = input_451_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_455_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_455_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_10_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99247168))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100295808))))[name = string("e_layers_10_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_10_feed_forward2_linear2_weight_to_fp16_lut, x = input_455_cast_fp16)[name = string("linear_99_cast_fp16")];
            fp16 var_5074_to_fp16 = const()[name = string("op_5074_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5075_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_5074_to_fp16)[name = string("op_5075_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_459_cast_fp16 = add(x = input_449_cast_fp16, y = var_5075_cast_fp16)[name = string("input_459_cast_fp16")];
            tensor<int32, [1]> input_461_axes_0 = const()[name = string("input_461_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_10_norm_out_weight_to_fp16 = const()[name = string("e_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100558016)))];
            tensor<fp16, [1024]> e_layers_10_norm_out_bias_to_fp16 = const()[name = string("e_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100560128)))];
            fp16 var_5079_to_fp16 = const()[name = string("op_5079_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_461_cast_fp16 = layer_norm(axes = input_461_axes_0, beta = e_layers_10_norm_out_bias_to_fp16, epsilon = var_5079_to_fp16, gamma = e_layers_10_norm_out_weight_to_fp16, x = input_459_cast_fp16)[name = string("input_461_cast_fp16")];
            tensor<int32, [1]> input_463_axes_0 = const()[name = string("input_463_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100562240)))];
            tensor<fp16, [1024]> e_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100564352)))];
            fp16 var_5086_to_fp16 = const()[name = string("op_5086_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_463_cast_fp16 = layer_norm(axes = input_463_axes_0, beta = e_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_5086_to_fp16, gamma = e_layers_11_norm_feed_forward1_weight_to_fp16, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_11_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100566464))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101615104))))[name = string("e_layers_11_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_11_feed_forward1_linear1_weight_to_fp16_lut, x = input_463_cast_fp16)[name = string("linear_100_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_467_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_467_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_11_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101877312))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102925952))))[name = string("e_layers_11_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_feed_forward1_linear2_weight_to_fp16_lut, x = input_467_cast_fp16)[name = string("linear_101_cast_fp16")];
            fp16 var_5103_to_fp16 = const()[name = string("op_5103_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5104_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_5103_to_fp16)[name = string("op_5104_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_471_cast_fp16 = add(x = input_461_cast_fp16, y = var_5104_cast_fp16)[name = string("input_471_cast_fp16")];
            tensor<int32, [1]> input_473_axes_0 = const()[name = string("input_473_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103188160)))];
            tensor<fp16, [1024]> e_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103190272)))];
            fp16 var_5108_to_fp16 = const()[name = string("op_5108_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_473_cast_fp16 = layer_norm(axes = input_473_axes_0, beta = e_layers_11_norm_self_att_bias_to_fp16, epsilon = var_5108_to_fp16, gamma = e_layers_11_norm_self_att_weight_to_fp16, x = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_11_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103192384))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103454592))))[name = string("e_layers_11_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_self_attn_q_proj_weight_to_fp16_lut, x = input_473_cast_fp16)[name = string("linear_102_cast_fp16")];
            tensor<int32, [4]> var_5121 = const()[name = string("op_5121"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5122_cast_fp16 = reshape(shape = var_5121, x = linear_102_cast_fp16)[name = string("op_5122_cast_fp16")];
            tensor<int32, [4]> q_23_perm_0 = const()[name = string("q_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_11_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103520192))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103782400))))[name = string("e_layers_11_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_self_attn_k_proj_weight_to_fp16_lut, x = input_473_cast_fp16)[name = string("linear_103_cast_fp16")];
            tensor<int32, [4]> var_5133 = const()[name = string("op_5133"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5134_cast_fp16 = reshape(shape = var_5133, x = linear_103_cast_fp16)[name = string("op_5134_cast_fp16")];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = string("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_11_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103848000))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104110208))))[name = string("e_layers_11_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_self_attn_v_proj_weight_to_fp16_lut, x = input_473_cast_fp16)[name = string("linear_104_cast_fp16")];
            tensor<int32, [4]> var_5145 = const()[name = string("op_5145"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5146_cast_fp16 = reshape(shape = var_5145, x = linear_104_cast_fp16)[name = string("op_5146_cast_fp16")];
            tensor<int32, [4]> v_23_perm_0 = const()[name = string("v_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_5170_to_fp16 = const()[name = string("op_5170_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104175808)))];
            tensor<fp16, [1, 8, 188, 128]> q_23_cast_fp16 = transpose(perm = q_23_perm_0, x = var_5122_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [1, 8, 188, 128]> var_5172_cast_fp16 = add(x = q_23_cast_fp16, y = var_5170_to_fp16)[name = string("op_5172_cast_fp16")];
            bool input_475_transpose_x_0 = const()[name = string("input_475_transpose_x_0"), val = bool(false)];
            bool input_475_transpose_y_0 = const()[name = string("input_475_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_23_to_fp16 = const()[name = string("p_23_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104177920)))];
            tensor<fp16, [1, 8, 188, 375]> input_475_cast_fp16 = matmul(transpose_x = input_475_transpose_x_0, transpose_y = input_475_transpose_y_0, x = var_5172_cast_fp16, y = p_23_to_fp16)[name = string("input_475_cast_fp16")];
            tensor<int32, [8]> var_5179_pad_0 = const()[name = string("op_5179_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5179_mode_0 = const()[name = string("op_5179_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5179_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_5179_mode_0, pad = var_5179_pad_0, x = input_475_cast_fp16)[name = string("op_5179_cast_fp16")];
            tensor<int32, [4]> var_5184 = const()[name = string("op_5184"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_5185_cast_fp16 = reshape(shape = var_5184, x = var_5179_cast_fp16)[name = string("op_5185_cast_fp16")];
            tensor<int32, [4]> var_5200_begin_0 = const()[name = string("op_5200_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5200_end_0 = const()[name = string("op_5200_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5200_end_mask_0 = const()[name = string("op_5200_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5200_cast_fp16 = slice_by_index(begin = var_5200_begin_0, end = var_5200_end_0, end_mask = var_5200_end_mask_0, x = var_5185_cast_fp16)[name = string("op_5200_cast_fp16")];
            tensor<int32, [4]> var_5205 = const()[name = string("op_5205"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5206_cast_fp16 = reshape(shape = var_5205, x = var_5200_cast_fp16)[name = string("op_5206_cast_fp16")];
            tensor<int32, [4]> bd_23_begin_0 = const()[name = string("bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_23_end_0 = const()[name = string("bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_23_end_mask_0 = const()[name = string("bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_23_cast_fp16 = slice_by_index(begin = bd_23_begin_0, end = bd_23_end_0, end_mask = bd_23_end_mask_0, x = var_5206_cast_fp16)[name = string("bd_23_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_5217_to_fp16 = const()[name = string("op_5217_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104945984)))];
            tensor<fp16, [1, 8, 188, 128]> var_5219_cast_fp16 = add(x = q_23_cast_fp16, y = var_5217_to_fp16)[name = string("op_5219_cast_fp16")];
            bool ac_23_transpose_x_0 = const()[name = string("ac_23_transpose_x_0"), val = bool(false)];
            bool ac_23_transpose_y_0 = const()[name = string("ac_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_59_cast_fp16 = transpose(perm = transpose_59_perm_0, x = var_5134_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [1, 8, 188, 188]> ac_23_cast_fp16 = matmul(transpose_x = ac_23_transpose_x_0, transpose_y = ac_23_transpose_y_0, x = var_5219_cast_fp16, y = transpose_59_cast_fp16)[name = string("ac_23_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5225_cast_fp16 = add(x = ac_23_cast_fp16, y = bd_23_cast_fp16)[name = string("op_5225_cast_fp16")];
            fp16 var_5226_to_fp16 = const()[name = string("op_5226_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5227_cast_fp16 = mul(x = var_5225_cast_fp16, y = var_5226_to_fp16)[name = string("op_5227_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5229_cast_fp16 = add(x = var_5227_cast_fp16, y = bias_15_cast_fp16)[name = string("op_5229_cast_fp16")];
            int32 var_5230 = const()[name = string("op_5230"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_23_cast_fp16 = softmax(axis = var_5230, x = var_5229_cast_fp16)[name = string("w_23_cast_fp16")];
            bool var_5233_transpose_x_0 = const()[name = string("op_5233_transpose_x_0"), val = bool(false)];
            bool var_5233_transpose_y_0 = const()[name = string("op_5233_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_23_cast_fp16 = transpose(perm = v_23_perm_0, x = var_5146_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [1, 8, 188, 128]> var_5233_cast_fp16 = matmul(transpose_x = var_5233_transpose_x_0, transpose_y = var_5233_transpose_y_0, x = w_23_cast_fp16, y = v_23_cast_fp16)[name = string("op_5233_cast_fp16")];
            tensor<int32, [4]> var_5236_perm_0 = const()[name = string("op_5236_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5240 = const()[name = string("op_5240"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5236_cast_fp16 = transpose(perm = var_5236_perm_0, x = var_5233_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [1, 188, 1024]> input_477_cast_fp16 = reshape(shape = var_5240, x = var_5236_cast_fp16)[name = string("input_477_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_11_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104948096))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105210304))))[name = string("e_layers_11_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_self_attn_o_proj_weight_to_fp16_lut, x = input_477_cast_fp16)[name = string("linear_106_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_479_cast_fp16 = add(x = input_471_cast_fp16, y = linear_106_cast_fp16)[name = string("input_479_cast_fp16")];
            tensor<int32, [1]> x_67_axes_0 = const()[name = string("x_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_11_norm_conv_weight_to_fp16 = const()[name = string("e_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105275904)))];
            tensor<fp16, [1024]> e_layers_11_norm_conv_bias_to_fp16 = const()[name = string("e_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105278016)))];
            fp16 var_5248_to_fp16 = const()[name = string("op_5248_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_67_cast_fp16 = layer_norm(axes = x_67_axes_0, beta = e_layers_11_norm_conv_bias_to_fp16, epsilon = var_5248_to_fp16, gamma = e_layers_11_norm_conv_weight_to_fp16, x = input_479_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_11_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105280128))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105804480))))[name = string("e_layers_11_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_5262_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_11_conv_pointwise_conv1_weight_to_fp16_lut, x = x_67_cast_fp16)[name = string("op_5262_cast_fp16")];
            tensor<int32, [3]> input_481_perm_0 = const()[name = string("input_481_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_5265_split_num_splits_0 = const()[name = string("op_5265_split_num_splits_0"), val = int32(2)];
            int32 var_5265_split_axis_0 = const()[name = string("op_5265_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_481_cast_fp16 = transpose(perm = input_481_perm_0, x = var_5262_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [1, 1024, 188]> var_5265_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_5265_split_cast_fp16_1 = split(axis = var_5265_split_axis_0, num_splits = var_5265_split_num_splits_0, x = input_481_cast_fp16)[name = string("op_5265_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5265_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5265_split_cast_fp16_1)[name = string("op_5265_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5265_cast_fp16 = mul(x = var_5265_split_cast_fp16_0, y = var_5265_split_1_sigmoid_cast_fp16)[name = string("op_5265_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_483_cast_fp16 = mul(x = var_5265_cast_fp16, y = var_2228_cast_fp16)[name = string("input_483_cast_fp16")];
            string input_485_pad_type_0 = const()[name = string("input_485_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_485_pad_0 = const()[name = string("input_485_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_485_groups_0 = const()[name = string("input_485_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_485_strides_0 = const()[name = string("input_485_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_485_dilations_0 = const()[name = string("input_485_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_63_to_fp16 = const()[name = string("const_63_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105935616)))];
            tensor<fp16, [1024]> const_64_to_fp16 = const()[name = string("const_64_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105954112)))];
            tensor<fp16, [1, 1024, 188]> input_487_cast_fp16 = conv(bias = const_64_to_fp16, dilations = input_485_dilations_0, groups = input_485_groups_0, pad = input_485_pad_0, pad_type = input_485_pad_type_0, strides = input_485_strides_0, weight = const_63_to_fp16, x = input_483_cast_fp16)[name = string("input_487_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_71_cast_fp16 = silu(x = input_487_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<int32, [3]> var_5305_perm_0 = const()[name = string("op_5305_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_11_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105956224))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106218432))))[name = string("e_layers_11_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_5305_cast_fp16 = transpose(perm = var_5305_perm_0, x = x_71_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [1, 188, 1024]> var_5307_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_conv_pointwise_conv2_weight_to_fp16_lut, x = var_5305_cast_fp16)[name = string("op_5307_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_489_cast_fp16 = add(x = input_479_cast_fp16, y = var_5307_cast_fp16)[name = string("input_489_cast_fp16")];
            tensor<int32, [1]> input_491_axes_0 = const()[name = string("input_491_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106284032)))];
            tensor<fp16, [1024]> e_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106286144)))];
            fp16 var_5315_to_fp16 = const()[name = string("op_5315_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_491_cast_fp16 = layer_norm(axes = input_491_axes_0, beta = e_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_5315_to_fp16, gamma = e_layers_11_norm_feed_forward2_weight_to_fp16, x = input_489_cast_fp16)[name = string("input_491_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_11_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106288256))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107336896))))[name = string("e_layers_11_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_11_feed_forward2_linear1_weight_to_fp16_lut, x = input_491_cast_fp16)[name = string("linear_107_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_495_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_495_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_11_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107599104))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108647744))))[name = string("e_layers_11_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_11_feed_forward2_linear2_weight_to_fp16_lut, x = input_495_cast_fp16)[name = string("linear_108_cast_fp16")];
            fp16 var_5332_to_fp16 = const()[name = string("op_5332_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5333_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_5332_to_fp16)[name = string("op_5333_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_499_cast_fp16 = add(x = input_489_cast_fp16, y = var_5333_cast_fp16)[name = string("input_499_cast_fp16")];
            tensor<int32, [1]> input_501_axes_0 = const()[name = string("input_501_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_11_norm_out_weight_to_fp16 = const()[name = string("e_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108909952)))];
            tensor<fp16, [1024]> e_layers_11_norm_out_bias_to_fp16 = const()[name = string("e_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108912064)))];
            fp16 var_5337_to_fp16 = const()[name = string("op_5337_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_501_cast_fp16 = layer_norm(axes = input_501_axes_0, beta = e_layers_11_norm_out_bias_to_fp16, epsilon = var_5337_to_fp16, gamma = e_layers_11_norm_out_weight_to_fp16, x = input_499_cast_fp16)[name = string("input_501_cast_fp16")];
            tensor<int32, [1]> input_503_axes_0 = const()[name = string("input_503_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108914176)))];
            tensor<fp16, [1024]> e_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108916288)))];
            fp16 var_5344_to_fp16 = const()[name = string("op_5344_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_503_cast_fp16 = layer_norm(axes = input_503_axes_0, beta = e_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_5344_to_fp16, gamma = e_layers_12_norm_feed_forward1_weight_to_fp16, x = input_501_cast_fp16)[name = string("input_503_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_12_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108918400))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109967040))))[name = string("e_layers_12_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_12_feed_forward1_linear1_weight_to_fp16_lut, x = input_503_cast_fp16)[name = string("linear_109_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_507_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_507_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_12_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110229248))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111277888))))[name = string("e_layers_12_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_feed_forward1_linear2_weight_to_fp16_lut, x = input_507_cast_fp16)[name = string("linear_110_cast_fp16")];
            fp16 var_5361_to_fp16 = const()[name = string("op_5361_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5362_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_5361_to_fp16)[name = string("op_5362_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_511_cast_fp16 = add(x = input_501_cast_fp16, y = var_5362_cast_fp16)[name = string("input_511_cast_fp16")];
            tensor<int32, [1]> input_513_axes_0 = const()[name = string("input_513_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111540096)))];
            tensor<fp16, [1024]> e_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111542208)))];
            fp16 var_5366_to_fp16 = const()[name = string("op_5366_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_513_cast_fp16 = layer_norm(axes = input_513_axes_0, beta = e_layers_12_norm_self_att_bias_to_fp16, epsilon = var_5366_to_fp16, gamma = e_layers_12_norm_self_att_weight_to_fp16, x = input_511_cast_fp16)[name = string("input_513_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_12_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111544320))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111806528))))[name = string("e_layers_12_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_self_attn_q_proj_weight_to_fp16_lut, x = input_513_cast_fp16)[name = string("linear_111_cast_fp16")];
            tensor<int32, [4]> var_5379 = const()[name = string("op_5379"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5380_cast_fp16 = reshape(shape = var_5379, x = linear_111_cast_fp16)[name = string("op_5380_cast_fp16")];
            tensor<int32, [4]> q_25_perm_0 = const()[name = string("q_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_12_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111872128))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112134336))))[name = string("e_layers_12_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_self_attn_k_proj_weight_to_fp16_lut, x = input_513_cast_fp16)[name = string("linear_112_cast_fp16")];
            tensor<int32, [4]> var_5391 = const()[name = string("op_5391"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5392_cast_fp16 = reshape(shape = var_5391, x = linear_112_cast_fp16)[name = string("op_5392_cast_fp16")];
            tensor<int32, [4]> transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_12_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112199936))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112462144))))[name = string("e_layers_12_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_self_attn_v_proj_weight_to_fp16_lut, x = input_513_cast_fp16)[name = string("linear_113_cast_fp16")];
            tensor<int32, [4]> var_5403 = const()[name = string("op_5403"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5404_cast_fp16 = reshape(shape = var_5403, x = linear_113_cast_fp16)[name = string("op_5404_cast_fp16")];
            tensor<int32, [4]> v_25_perm_0 = const()[name = string("v_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_5428_to_fp16 = const()[name = string("op_5428_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112527744)))];
            tensor<fp16, [1, 8, 188, 128]> q_25_cast_fp16 = transpose(perm = q_25_perm_0, x = var_5380_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [1, 8, 188, 128]> var_5430_cast_fp16 = add(x = q_25_cast_fp16, y = var_5428_to_fp16)[name = string("op_5430_cast_fp16")];
            bool input_515_transpose_x_0 = const()[name = string("input_515_transpose_x_0"), val = bool(false)];
            bool input_515_transpose_y_0 = const()[name = string("input_515_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_25_to_fp16 = const()[name = string("p_25_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112529856)))];
            tensor<fp16, [1, 8, 188, 375]> input_515_cast_fp16 = matmul(transpose_x = input_515_transpose_x_0, transpose_y = input_515_transpose_y_0, x = var_5430_cast_fp16, y = p_25_to_fp16)[name = string("input_515_cast_fp16")];
            tensor<int32, [8]> var_5437_pad_0 = const()[name = string("op_5437_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5437_mode_0 = const()[name = string("op_5437_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5437_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = var_5437_mode_0, pad = var_5437_pad_0, x = input_515_cast_fp16)[name = string("op_5437_cast_fp16")];
            tensor<int32, [4]> var_5442 = const()[name = string("op_5442"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_5443_cast_fp16 = reshape(shape = var_5442, x = var_5437_cast_fp16)[name = string("op_5443_cast_fp16")];
            tensor<int32, [4]> var_5458_begin_0 = const()[name = string("op_5458_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5458_end_0 = const()[name = string("op_5458_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5458_end_mask_0 = const()[name = string("op_5458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5458_cast_fp16 = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = var_5443_cast_fp16)[name = string("op_5458_cast_fp16")];
            tensor<int32, [4]> var_5463 = const()[name = string("op_5463"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5464_cast_fp16 = reshape(shape = var_5463, x = var_5458_cast_fp16)[name = string("op_5464_cast_fp16")];
            tensor<int32, [4]> bd_25_begin_0 = const()[name = string("bd_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_25_end_0 = const()[name = string("bd_25_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_25_end_mask_0 = const()[name = string("bd_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_25_cast_fp16 = slice_by_index(begin = bd_25_begin_0, end = bd_25_end_0, end_mask = bd_25_end_mask_0, x = var_5464_cast_fp16)[name = string("bd_25_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_5475_to_fp16 = const()[name = string("op_5475_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113297920)))];
            tensor<fp16, [1, 8, 188, 128]> var_5477_cast_fp16 = add(x = q_25_cast_fp16, y = var_5475_to_fp16)[name = string("op_5477_cast_fp16")];
            bool ac_25_transpose_x_0 = const()[name = string("ac_25_transpose_x_0"), val = bool(false)];
            bool ac_25_transpose_y_0 = const()[name = string("ac_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_60_cast_fp16 = transpose(perm = transpose_60_perm_0, x = var_5392_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [1, 8, 188, 188]> ac_25_cast_fp16 = matmul(transpose_x = ac_25_transpose_x_0, transpose_y = ac_25_transpose_y_0, x = var_5477_cast_fp16, y = transpose_60_cast_fp16)[name = string("ac_25_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5483_cast_fp16 = add(x = ac_25_cast_fp16, y = bd_25_cast_fp16)[name = string("op_5483_cast_fp16")];
            fp16 var_5484_to_fp16 = const()[name = string("op_5484_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5485_cast_fp16 = mul(x = var_5483_cast_fp16, y = var_5484_to_fp16)[name = string("op_5485_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5487_cast_fp16 = add(x = var_5485_cast_fp16, y = bias_15_cast_fp16)[name = string("op_5487_cast_fp16")];
            int32 var_5488 = const()[name = string("op_5488"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_25_cast_fp16 = softmax(axis = var_5488, x = var_5487_cast_fp16)[name = string("w_25_cast_fp16")];
            bool var_5491_transpose_x_0 = const()[name = string("op_5491_transpose_x_0"), val = bool(false)];
            bool var_5491_transpose_y_0 = const()[name = string("op_5491_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_25_cast_fp16 = transpose(perm = v_25_perm_0, x = var_5404_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [1, 8, 188, 128]> var_5491_cast_fp16 = matmul(transpose_x = var_5491_transpose_x_0, transpose_y = var_5491_transpose_y_0, x = w_25_cast_fp16, y = v_25_cast_fp16)[name = string("op_5491_cast_fp16")];
            tensor<int32, [4]> var_5494_perm_0 = const()[name = string("op_5494_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5498 = const()[name = string("op_5498"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5494_cast_fp16 = transpose(perm = var_5494_perm_0, x = var_5491_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [1, 188, 1024]> input_517_cast_fp16 = reshape(shape = var_5498, x = var_5494_cast_fp16)[name = string("input_517_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_12_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113300032))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113562240))))[name = string("e_layers_12_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_self_attn_o_proj_weight_to_fp16_lut, x = input_517_cast_fp16)[name = string("linear_115_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_519_cast_fp16 = add(x = input_511_cast_fp16, y = linear_115_cast_fp16)[name = string("input_519_cast_fp16")];
            tensor<int32, [1]> x_73_axes_0 = const()[name = string("x_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_12_norm_conv_weight_to_fp16 = const()[name = string("e_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113627840)))];
            tensor<fp16, [1024]> e_layers_12_norm_conv_bias_to_fp16 = const()[name = string("e_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113629952)))];
            fp16 var_5506_to_fp16 = const()[name = string("op_5506_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_73_cast_fp16 = layer_norm(axes = x_73_axes_0, beta = e_layers_12_norm_conv_bias_to_fp16, epsilon = var_5506_to_fp16, gamma = e_layers_12_norm_conv_weight_to_fp16, x = input_519_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_12_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113632064))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114156416))))[name = string("e_layers_12_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_5520_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_12_conv_pointwise_conv1_weight_to_fp16_lut, x = x_73_cast_fp16)[name = string("op_5520_cast_fp16")];
            tensor<int32, [3]> input_521_perm_0 = const()[name = string("input_521_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_5523_split_num_splits_0 = const()[name = string("op_5523_split_num_splits_0"), val = int32(2)];
            int32 var_5523_split_axis_0 = const()[name = string("op_5523_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_521_cast_fp16 = transpose(perm = input_521_perm_0, x = var_5520_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [1, 1024, 188]> var_5523_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_5523_split_cast_fp16_1 = split(axis = var_5523_split_axis_0, num_splits = var_5523_split_num_splits_0, x = input_521_cast_fp16)[name = string("op_5523_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5523_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5523_split_cast_fp16_1)[name = string("op_5523_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5523_cast_fp16 = mul(x = var_5523_split_cast_fp16_0, y = var_5523_split_1_sigmoid_cast_fp16)[name = string("op_5523_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_523_cast_fp16 = mul(x = var_5523_cast_fp16, y = var_2228_cast_fp16)[name = string("input_523_cast_fp16")];
            string input_525_pad_type_0 = const()[name = string("input_525_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_525_pad_0 = const()[name = string("input_525_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_525_groups_0 = const()[name = string("input_525_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_525_strides_0 = const()[name = string("input_525_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_525_dilations_0 = const()[name = string("input_525_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_65_to_fp16 = const()[name = string("const_65_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114287552)))];
            tensor<fp16, [1024]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114306048)))];
            tensor<fp16, [1, 1024, 188]> input_527_cast_fp16 = conv(bias = const_66_to_fp16, dilations = input_525_dilations_0, groups = input_525_groups_0, pad = input_525_pad_0, pad_type = input_525_pad_type_0, strides = input_525_strides_0, weight = const_65_to_fp16, x = input_523_cast_fp16)[name = string("input_527_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_77_cast_fp16 = silu(x = input_527_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<int32, [3]> var_5563_perm_0 = const()[name = string("op_5563_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_12_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114308160))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114570368))))[name = string("e_layers_12_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_5563_cast_fp16 = transpose(perm = var_5563_perm_0, x = x_77_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [1, 188, 1024]> var_5565_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_conv_pointwise_conv2_weight_to_fp16_lut, x = var_5563_cast_fp16)[name = string("op_5565_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_529_cast_fp16 = add(x = input_519_cast_fp16, y = var_5565_cast_fp16)[name = string("input_529_cast_fp16")];
            tensor<int32, [1]> input_531_axes_0 = const()[name = string("input_531_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114635968)))];
            tensor<fp16, [1024]> e_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114638080)))];
            fp16 var_5573_to_fp16 = const()[name = string("op_5573_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_531_cast_fp16 = layer_norm(axes = input_531_axes_0, beta = e_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_5573_to_fp16, gamma = e_layers_12_norm_feed_forward2_weight_to_fp16, x = input_529_cast_fp16)[name = string("input_531_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_12_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114640192))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115688832))))[name = string("e_layers_12_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_12_feed_forward2_linear1_weight_to_fp16_lut, x = input_531_cast_fp16)[name = string("linear_116_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_535_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_535_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_12_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115951040))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116999680))))[name = string("e_layers_12_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_12_feed_forward2_linear2_weight_to_fp16_lut, x = input_535_cast_fp16)[name = string("linear_117_cast_fp16")];
            fp16 var_5590_to_fp16 = const()[name = string("op_5590_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5591_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_5590_to_fp16)[name = string("op_5591_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_539_cast_fp16 = add(x = input_529_cast_fp16, y = var_5591_cast_fp16)[name = string("input_539_cast_fp16")];
            tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_12_norm_out_weight_to_fp16 = const()[name = string("e_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117261888)))];
            tensor<fp16, [1024]> e_layers_12_norm_out_bias_to_fp16 = const()[name = string("e_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117264000)))];
            fp16 var_5595_to_fp16 = const()[name = string("op_5595_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = e_layers_12_norm_out_bias_to_fp16, epsilon = var_5595_to_fp16, gamma = e_layers_12_norm_out_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
            tensor<int32, [1]> input_543_axes_0 = const()[name = string("input_543_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117266112)))];
            tensor<fp16, [1024]> e_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117268224)))];
            fp16 var_5602_to_fp16 = const()[name = string("op_5602_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_543_cast_fp16 = layer_norm(axes = input_543_axes_0, beta = e_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_5602_to_fp16, gamma = e_layers_13_norm_feed_forward1_weight_to_fp16, x = input_541_cast_fp16)[name = string("input_543_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_13_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117270336))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118318976))))[name = string("e_layers_13_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_13_feed_forward1_linear1_weight_to_fp16_lut, x = input_543_cast_fp16)[name = string("linear_118_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_547_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_547_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_13_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118581184))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119629824))))[name = string("e_layers_13_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_feed_forward1_linear2_weight_to_fp16_lut, x = input_547_cast_fp16)[name = string("linear_119_cast_fp16")];
            fp16 var_5619_to_fp16 = const()[name = string("op_5619_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5620_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_5619_to_fp16)[name = string("op_5620_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_551_cast_fp16 = add(x = input_541_cast_fp16, y = var_5620_cast_fp16)[name = string("input_551_cast_fp16")];
            tensor<int32, [1]> input_553_axes_0 = const()[name = string("input_553_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119892032)))];
            tensor<fp16, [1024]> e_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119894144)))];
            fp16 var_5624_to_fp16 = const()[name = string("op_5624_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_553_cast_fp16 = layer_norm(axes = input_553_axes_0, beta = e_layers_13_norm_self_att_bias_to_fp16, epsilon = var_5624_to_fp16, gamma = e_layers_13_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("input_553_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_13_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119896256))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120158464))))[name = string("e_layers_13_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_self_attn_q_proj_weight_to_fp16_lut, x = input_553_cast_fp16)[name = string("linear_120_cast_fp16")];
            tensor<int32, [4]> var_5637 = const()[name = string("op_5637"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5638_cast_fp16 = reshape(shape = var_5637, x = linear_120_cast_fp16)[name = string("op_5638_cast_fp16")];
            tensor<int32, [4]> q_27_perm_0 = const()[name = string("q_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_13_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120224064))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120486272))))[name = string("e_layers_13_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_self_attn_k_proj_weight_to_fp16_lut, x = input_553_cast_fp16)[name = string("linear_121_cast_fp16")];
            tensor<int32, [4]> var_5649 = const()[name = string("op_5649"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5650_cast_fp16 = reshape(shape = var_5649, x = linear_121_cast_fp16)[name = string("op_5650_cast_fp16")];
            tensor<int32, [4]> transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_13_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120551872))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120814080))))[name = string("e_layers_13_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_self_attn_v_proj_weight_to_fp16_lut, x = input_553_cast_fp16)[name = string("linear_122_cast_fp16")];
            tensor<int32, [4]> var_5661 = const()[name = string("op_5661"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5662_cast_fp16 = reshape(shape = var_5661, x = linear_122_cast_fp16)[name = string("op_5662_cast_fp16")];
            tensor<int32, [4]> v_27_perm_0 = const()[name = string("v_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_5686_to_fp16 = const()[name = string("op_5686_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120879680)))];
            tensor<fp16, [1, 8, 188, 128]> q_27_cast_fp16 = transpose(perm = q_27_perm_0, x = var_5638_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [1, 8, 188, 128]> var_5688_cast_fp16 = add(x = q_27_cast_fp16, y = var_5686_to_fp16)[name = string("op_5688_cast_fp16")];
            bool input_555_transpose_x_0 = const()[name = string("input_555_transpose_x_0"), val = bool(false)];
            bool input_555_transpose_y_0 = const()[name = string("input_555_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_27_to_fp16 = const()[name = string("p_27_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120881792)))];
            tensor<fp16, [1, 8, 188, 375]> input_555_cast_fp16 = matmul(transpose_x = input_555_transpose_x_0, transpose_y = input_555_transpose_y_0, x = var_5688_cast_fp16, y = p_27_to_fp16)[name = string("input_555_cast_fp16")];
            tensor<int32, [8]> var_5695_pad_0 = const()[name = string("op_5695_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5695_mode_0 = const()[name = string("op_5695_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5695_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = var_5695_mode_0, pad = var_5695_pad_0, x = input_555_cast_fp16)[name = string("op_5695_cast_fp16")];
            tensor<int32, [4]> var_5700 = const()[name = string("op_5700"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_5701_cast_fp16 = reshape(shape = var_5700, x = var_5695_cast_fp16)[name = string("op_5701_cast_fp16")];
            tensor<int32, [4]> var_5716_begin_0 = const()[name = string("op_5716_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5716_end_0 = const()[name = string("op_5716_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5716_end_mask_0 = const()[name = string("op_5716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5716_cast_fp16 = slice_by_index(begin = var_5716_begin_0, end = var_5716_end_0, end_mask = var_5716_end_mask_0, x = var_5701_cast_fp16)[name = string("op_5716_cast_fp16")];
            tensor<int32, [4]> var_5721 = const()[name = string("op_5721"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5722_cast_fp16 = reshape(shape = var_5721, x = var_5716_cast_fp16)[name = string("op_5722_cast_fp16")];
            tensor<int32, [4]> bd_27_begin_0 = const()[name = string("bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_27_end_0 = const()[name = string("bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_27_end_mask_0 = const()[name = string("bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_27_cast_fp16 = slice_by_index(begin = bd_27_begin_0, end = bd_27_end_0, end_mask = bd_27_end_mask_0, x = var_5722_cast_fp16)[name = string("bd_27_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_5733_to_fp16 = const()[name = string("op_5733_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121649856)))];
            tensor<fp16, [1, 8, 188, 128]> var_5735_cast_fp16 = add(x = q_27_cast_fp16, y = var_5733_to_fp16)[name = string("op_5735_cast_fp16")];
            bool ac_27_transpose_x_0 = const()[name = string("ac_27_transpose_x_0"), val = bool(false)];
            bool ac_27_transpose_y_0 = const()[name = string("ac_27_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_61_cast_fp16 = transpose(perm = transpose_61_perm_0, x = var_5650_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [1, 8, 188, 188]> ac_27_cast_fp16 = matmul(transpose_x = ac_27_transpose_x_0, transpose_y = ac_27_transpose_y_0, x = var_5735_cast_fp16, y = transpose_61_cast_fp16)[name = string("ac_27_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5741_cast_fp16 = add(x = ac_27_cast_fp16, y = bd_27_cast_fp16)[name = string("op_5741_cast_fp16")];
            fp16 var_5742_to_fp16 = const()[name = string("op_5742_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_5743_cast_fp16 = mul(x = var_5741_cast_fp16, y = var_5742_to_fp16)[name = string("op_5743_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5745_cast_fp16 = add(x = var_5743_cast_fp16, y = bias_15_cast_fp16)[name = string("op_5745_cast_fp16")];
            int32 var_5746 = const()[name = string("op_5746"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_27_cast_fp16 = softmax(axis = var_5746, x = var_5745_cast_fp16)[name = string("w_27_cast_fp16")];
            bool var_5749_transpose_x_0 = const()[name = string("op_5749_transpose_x_0"), val = bool(false)];
            bool var_5749_transpose_y_0 = const()[name = string("op_5749_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = var_5662_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [1, 8, 188, 128]> var_5749_cast_fp16 = matmul(transpose_x = var_5749_transpose_x_0, transpose_y = var_5749_transpose_y_0, x = w_27_cast_fp16, y = v_27_cast_fp16)[name = string("op_5749_cast_fp16")];
            tensor<int32, [4]> var_5752_perm_0 = const()[name = string("op_5752_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5756 = const()[name = string("op_5756"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5752_cast_fp16 = transpose(perm = var_5752_perm_0, x = var_5749_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [1, 188, 1024]> input_557_cast_fp16 = reshape(shape = var_5756, x = var_5752_cast_fp16)[name = string("input_557_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_13_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121651968))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121914176))))[name = string("e_layers_13_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_self_attn_o_proj_weight_to_fp16_lut, x = input_557_cast_fp16)[name = string("linear_124_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_124_cast_fp16)[name = string("input_559_cast_fp16")];
            tensor<int32, [1]> x_79_axes_0 = const()[name = string("x_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_13_norm_conv_weight_to_fp16 = const()[name = string("e_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121979776)))];
            tensor<fp16, [1024]> e_layers_13_norm_conv_bias_to_fp16 = const()[name = string("e_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121981888)))];
            fp16 var_5764_to_fp16 = const()[name = string("op_5764_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_79_cast_fp16 = layer_norm(axes = x_79_axes_0, beta = e_layers_13_norm_conv_bias_to_fp16, epsilon = var_5764_to_fp16, gamma = e_layers_13_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_13_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121984000))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122508352))))[name = string("e_layers_13_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_5778_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_13_conv_pointwise_conv1_weight_to_fp16_lut, x = x_79_cast_fp16)[name = string("op_5778_cast_fp16")];
            tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_5781_split_num_splits_0 = const()[name = string("op_5781_split_num_splits_0"), val = int32(2)];
            int32 var_5781_split_axis_0 = const()[name = string("op_5781_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = var_5778_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 1024, 188]> var_5781_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_5781_split_cast_fp16_1 = split(axis = var_5781_split_axis_0, num_splits = var_5781_split_num_splits_0, x = input_561_cast_fp16)[name = string("op_5781_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5781_split_1_sigmoid_cast_fp16 = sigmoid(x = var_5781_split_cast_fp16_1)[name = string("op_5781_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_5781_cast_fp16 = mul(x = var_5781_split_cast_fp16_0, y = var_5781_split_1_sigmoid_cast_fp16)[name = string("op_5781_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_563_cast_fp16 = mul(x = var_5781_cast_fp16, y = var_2228_cast_fp16)[name = string("input_563_cast_fp16")];
            string input_565_pad_type_0 = const()[name = string("input_565_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_565_pad_0 = const()[name = string("input_565_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_565_groups_0 = const()[name = string("input_565_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_565_strides_0 = const()[name = string("input_565_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_565_dilations_0 = const()[name = string("input_565_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122639488)))];
            tensor<fp16, [1024]> const_68_to_fp16 = const()[name = string("const_68_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122657984)))];
            tensor<fp16, [1, 1024, 188]> input_567_cast_fp16 = conv(bias = const_68_to_fp16, dilations = input_565_dilations_0, groups = input_565_groups_0, pad = input_565_pad_0, pad_type = input_565_pad_type_0, strides = input_565_strides_0, weight = const_67_to_fp16, x = input_563_cast_fp16)[name = string("input_567_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_83_cast_fp16 = silu(x = input_567_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<int32, [3]> var_5821_perm_0 = const()[name = string("op_5821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_13_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122660096))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122922304))))[name = string("e_layers_13_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_5821_cast_fp16 = transpose(perm = var_5821_perm_0, x = x_83_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 188, 1024]> var_5823_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_conv_pointwise_conv2_weight_to_fp16_lut, x = var_5821_cast_fp16)[name = string("op_5823_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_569_cast_fp16 = add(x = input_559_cast_fp16, y = var_5823_cast_fp16)[name = string("input_569_cast_fp16")];
            tensor<int32, [1]> input_571_axes_0 = const()[name = string("input_571_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122987904)))];
            tensor<fp16, [1024]> e_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122990016)))];
            fp16 var_5831_to_fp16 = const()[name = string("op_5831_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_571_cast_fp16 = layer_norm(axes = input_571_axes_0, beta = e_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_5831_to_fp16, gamma = e_layers_13_norm_feed_forward2_weight_to_fp16, x = input_569_cast_fp16)[name = string("input_571_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_13_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122992128))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124040768))))[name = string("e_layers_13_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_13_feed_forward2_linear1_weight_to_fp16_lut, x = input_571_cast_fp16)[name = string("linear_125_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_575_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_575_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_13_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124302976))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125351616))))[name = string("e_layers_13_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_13_feed_forward2_linear2_weight_to_fp16_lut, x = input_575_cast_fp16)[name = string("linear_126_cast_fp16")];
            fp16 var_5848_to_fp16 = const()[name = string("op_5848_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5849_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_5848_to_fp16)[name = string("op_5849_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_579_cast_fp16 = add(x = input_569_cast_fp16, y = var_5849_cast_fp16)[name = string("input_579_cast_fp16")];
            tensor<int32, [1]> input_581_axes_0 = const()[name = string("input_581_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_13_norm_out_weight_to_fp16 = const()[name = string("e_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125613824)))];
            tensor<fp16, [1024]> e_layers_13_norm_out_bias_to_fp16 = const()[name = string("e_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125615936)))];
            fp16 var_5853_to_fp16 = const()[name = string("op_5853_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_581_cast_fp16 = layer_norm(axes = input_581_axes_0, beta = e_layers_13_norm_out_bias_to_fp16, epsilon = var_5853_to_fp16, gamma = e_layers_13_norm_out_weight_to_fp16, x = input_579_cast_fp16)[name = string("input_581_cast_fp16")];
            tensor<int32, [1]> input_583_axes_0 = const()[name = string("input_583_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125618048)))];
            tensor<fp16, [1024]> e_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125620160)))];
            fp16 var_5860_to_fp16 = const()[name = string("op_5860_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_583_cast_fp16 = layer_norm(axes = input_583_axes_0, beta = e_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_5860_to_fp16, gamma = e_layers_14_norm_feed_forward1_weight_to_fp16, x = input_581_cast_fp16)[name = string("input_583_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_14_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125622272))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126670912))))[name = string("e_layers_14_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_14_feed_forward1_linear1_weight_to_fp16_lut, x = input_583_cast_fp16)[name = string("linear_127_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_587_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_587_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_14_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126933120))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127981760))))[name = string("e_layers_14_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_feed_forward1_linear2_weight_to_fp16_lut, x = input_587_cast_fp16)[name = string("linear_128_cast_fp16")];
            fp16 var_5877_to_fp16 = const()[name = string("op_5877_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5878_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_5877_to_fp16)[name = string("op_5878_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_591_cast_fp16 = add(x = input_581_cast_fp16, y = var_5878_cast_fp16)[name = string("input_591_cast_fp16")];
            tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128243968)))];
            tensor<fp16, [1024]> e_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128246080)))];
            fp16 var_5882_to_fp16 = const()[name = string("op_5882_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = e_layers_14_norm_self_att_bias_to_fp16, epsilon = var_5882_to_fp16, gamma = e_layers_14_norm_self_att_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_14_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128248192))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128510400))))[name = string("e_layers_14_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_self_attn_q_proj_weight_to_fp16_lut, x = input_593_cast_fp16)[name = string("linear_129_cast_fp16")];
            tensor<int32, [4]> var_5895 = const()[name = string("op_5895"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5896_cast_fp16 = reshape(shape = var_5895, x = linear_129_cast_fp16)[name = string("op_5896_cast_fp16")];
            tensor<int32, [4]> q_29_perm_0 = const()[name = string("q_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_14_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128576000))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128838208))))[name = string("e_layers_14_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_self_attn_k_proj_weight_to_fp16_lut, x = input_593_cast_fp16)[name = string("linear_130_cast_fp16")];
            tensor<int32, [4]> var_5907 = const()[name = string("op_5907"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5908_cast_fp16 = reshape(shape = var_5907, x = linear_130_cast_fp16)[name = string("op_5908_cast_fp16")];
            tensor<int32, [4]> transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_14_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128903808))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129166016))))[name = string("e_layers_14_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_self_attn_v_proj_weight_to_fp16_lut, x = input_593_cast_fp16)[name = string("linear_131_cast_fp16")];
            tensor<int32, [4]> var_5919 = const()[name = string("op_5919"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_5920_cast_fp16 = reshape(shape = var_5919, x = linear_131_cast_fp16)[name = string("op_5920_cast_fp16")];
            tensor<int32, [4]> v_29_perm_0 = const()[name = string("v_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_5944_to_fp16 = const()[name = string("op_5944_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129231616)))];
            tensor<fp16, [1, 8, 188, 128]> q_29_cast_fp16 = transpose(perm = q_29_perm_0, x = var_5896_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 8, 188, 128]> var_5946_cast_fp16 = add(x = q_29_cast_fp16, y = var_5944_to_fp16)[name = string("op_5946_cast_fp16")];
            bool input_595_transpose_x_0 = const()[name = string("input_595_transpose_x_0"), val = bool(false)];
            bool input_595_transpose_y_0 = const()[name = string("input_595_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_29_to_fp16 = const()[name = string("p_29_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129233728)))];
            tensor<fp16, [1, 8, 188, 375]> input_595_cast_fp16 = matmul(transpose_x = input_595_transpose_x_0, transpose_y = input_595_transpose_y_0, x = var_5946_cast_fp16, y = p_29_to_fp16)[name = string("input_595_cast_fp16")];
            tensor<int32, [8]> var_5953_pad_0 = const()[name = string("op_5953_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_5953_mode_0 = const()[name = string("op_5953_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_5953_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_5953_mode_0, pad = var_5953_pad_0, x = input_595_cast_fp16)[name = string("op_5953_cast_fp16")];
            tensor<int32, [4]> var_5958 = const()[name = string("op_5958"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_5959_cast_fp16 = reshape(shape = var_5958, x = var_5953_cast_fp16)[name = string("op_5959_cast_fp16")];
            tensor<int32, [4]> var_5974_begin_0 = const()[name = string("op_5974_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5974_end_0 = const()[name = string("op_5974_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5974_end_mask_0 = const()[name = string("op_5974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5974_cast_fp16 = slice_by_index(begin = var_5974_begin_0, end = var_5974_end_0, end_mask = var_5974_end_mask_0, x = var_5959_cast_fp16)[name = string("op_5974_cast_fp16")];
            tensor<int32, [4]> var_5979 = const()[name = string("op_5979"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_5980_cast_fp16 = reshape(shape = var_5979, x = var_5974_cast_fp16)[name = string("op_5980_cast_fp16")];
            tensor<int32, [4]> bd_29_begin_0 = const()[name = string("bd_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_29_end_0 = const()[name = string("bd_29_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_29_end_mask_0 = const()[name = string("bd_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_29_cast_fp16 = slice_by_index(begin = bd_29_begin_0, end = bd_29_end_0, end_mask = bd_29_end_mask_0, x = var_5980_cast_fp16)[name = string("bd_29_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_5991_to_fp16 = const()[name = string("op_5991_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130001792)))];
            tensor<fp16, [1, 8, 188, 128]> var_5993_cast_fp16 = add(x = q_29_cast_fp16, y = var_5991_to_fp16)[name = string("op_5993_cast_fp16")];
            bool ac_29_transpose_x_0 = const()[name = string("ac_29_transpose_x_0"), val = bool(false)];
            bool ac_29_transpose_y_0 = const()[name = string("ac_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_62_cast_fp16 = transpose(perm = transpose_62_perm_0, x = var_5908_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 8, 188, 188]> ac_29_cast_fp16 = matmul(transpose_x = ac_29_transpose_x_0, transpose_y = ac_29_transpose_y_0, x = var_5993_cast_fp16, y = transpose_62_cast_fp16)[name = string("ac_29_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5999_cast_fp16 = add(x = ac_29_cast_fp16, y = bd_29_cast_fp16)[name = string("op_5999_cast_fp16")];
            fp16 var_6000_to_fp16 = const()[name = string("op_6000_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6001_cast_fp16 = mul(x = var_5999_cast_fp16, y = var_6000_to_fp16)[name = string("op_6001_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6003_cast_fp16 = add(x = var_6001_cast_fp16, y = bias_15_cast_fp16)[name = string("op_6003_cast_fp16")];
            int32 var_6004 = const()[name = string("op_6004"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_29_cast_fp16 = softmax(axis = var_6004, x = var_6003_cast_fp16)[name = string("w_29_cast_fp16")];
            bool var_6007_transpose_x_0 = const()[name = string("op_6007_transpose_x_0"), val = bool(false)];
            bool var_6007_transpose_y_0 = const()[name = string("op_6007_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_29_cast_fp16 = transpose(perm = v_29_perm_0, x = var_5920_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 8, 188, 128]> var_6007_cast_fp16 = matmul(transpose_x = var_6007_transpose_x_0, transpose_y = var_6007_transpose_y_0, x = w_29_cast_fp16, y = v_29_cast_fp16)[name = string("op_6007_cast_fp16")];
            tensor<int32, [4]> var_6010_perm_0 = const()[name = string("op_6010_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6014 = const()[name = string("op_6014"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_6010_cast_fp16 = transpose(perm = var_6010_perm_0, x = var_6007_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 188, 1024]> input_597_cast_fp16 = reshape(shape = var_6014, x = var_6010_cast_fp16)[name = string("input_597_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_14_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130003904))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130266112))))[name = string("e_layers_14_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_self_attn_o_proj_weight_to_fp16_lut, x = input_597_cast_fp16)[name = string("linear_133_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_599_cast_fp16 = add(x = input_591_cast_fp16, y = linear_133_cast_fp16)[name = string("input_599_cast_fp16")];
            tensor<int32, [1]> x_85_axes_0 = const()[name = string("x_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_14_norm_conv_weight_to_fp16 = const()[name = string("e_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130331712)))];
            tensor<fp16, [1024]> e_layers_14_norm_conv_bias_to_fp16 = const()[name = string("e_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130333824)))];
            fp16 var_6022_to_fp16 = const()[name = string("op_6022_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_85_cast_fp16 = layer_norm(axes = x_85_axes_0, beta = e_layers_14_norm_conv_bias_to_fp16, epsilon = var_6022_to_fp16, gamma = e_layers_14_norm_conv_weight_to_fp16, x = input_599_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_14_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130335936))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130860288))))[name = string("e_layers_14_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_6036_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_14_conv_pointwise_conv1_weight_to_fp16_lut, x = x_85_cast_fp16)[name = string("op_6036_cast_fp16")];
            tensor<int32, [3]> input_601_perm_0 = const()[name = string("input_601_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_6039_split_num_splits_0 = const()[name = string("op_6039_split_num_splits_0"), val = int32(2)];
            int32 var_6039_split_axis_0 = const()[name = string("op_6039_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_601_cast_fp16 = transpose(perm = input_601_perm_0, x = var_6036_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 1024, 188]> var_6039_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_6039_split_cast_fp16_1 = split(axis = var_6039_split_axis_0, num_splits = var_6039_split_num_splits_0, x = input_601_cast_fp16)[name = string("op_6039_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6039_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6039_split_cast_fp16_1)[name = string("op_6039_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6039_cast_fp16 = mul(x = var_6039_split_cast_fp16_0, y = var_6039_split_1_sigmoid_cast_fp16)[name = string("op_6039_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_603_cast_fp16 = mul(x = var_6039_cast_fp16, y = var_2228_cast_fp16)[name = string("input_603_cast_fp16")];
            string input_605_pad_type_0 = const()[name = string("input_605_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_605_pad_0 = const()[name = string("input_605_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_605_groups_0 = const()[name = string("input_605_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_605_strides_0 = const()[name = string("input_605_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_605_dilations_0 = const()[name = string("input_605_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130991424)))];
            tensor<fp16, [1024]> const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131009920)))];
            tensor<fp16, [1, 1024, 188]> input_607_cast_fp16 = conv(bias = const_70_to_fp16, dilations = input_605_dilations_0, groups = input_605_groups_0, pad = input_605_pad_0, pad_type = input_605_pad_type_0, strides = input_605_strides_0, weight = const_69_to_fp16, x = input_603_cast_fp16)[name = string("input_607_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_89_cast_fp16 = silu(x = input_607_cast_fp16)[name = string("x_89_cast_fp16")];
            tensor<int32, [3]> var_6079_perm_0 = const()[name = string("op_6079_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_14_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131012032))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131274240))))[name = string("e_layers_14_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_6079_cast_fp16 = transpose(perm = var_6079_perm_0, x = x_89_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 188, 1024]> var_6081_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_conv_pointwise_conv2_weight_to_fp16_lut, x = var_6079_cast_fp16)[name = string("op_6081_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_609_cast_fp16 = add(x = input_599_cast_fp16, y = var_6081_cast_fp16)[name = string("input_609_cast_fp16")];
            tensor<int32, [1]> input_611_axes_0 = const()[name = string("input_611_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131339840)))];
            tensor<fp16, [1024]> e_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131341952)))];
            fp16 var_6089_to_fp16 = const()[name = string("op_6089_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_611_cast_fp16 = layer_norm(axes = input_611_axes_0, beta = e_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_6089_to_fp16, gamma = e_layers_14_norm_feed_forward2_weight_to_fp16, x = input_609_cast_fp16)[name = string("input_611_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_14_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131344064))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132392704))))[name = string("e_layers_14_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_14_feed_forward2_linear1_weight_to_fp16_lut, x = input_611_cast_fp16)[name = string("linear_134_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_615_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_615_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_14_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132654912))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133703552))))[name = string("e_layers_14_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_14_feed_forward2_linear2_weight_to_fp16_lut, x = input_615_cast_fp16)[name = string("linear_135_cast_fp16")];
            fp16 var_6106_to_fp16 = const()[name = string("op_6106_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6107_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_6106_to_fp16)[name = string("op_6107_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_619_cast_fp16 = add(x = input_609_cast_fp16, y = var_6107_cast_fp16)[name = string("input_619_cast_fp16")];
            tensor<int32, [1]> input_621_axes_0 = const()[name = string("input_621_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_14_norm_out_weight_to_fp16 = const()[name = string("e_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133965760)))];
            tensor<fp16, [1024]> e_layers_14_norm_out_bias_to_fp16 = const()[name = string("e_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133967872)))];
            fp16 var_6111_to_fp16 = const()[name = string("op_6111_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_621_cast_fp16 = layer_norm(axes = input_621_axes_0, beta = e_layers_14_norm_out_bias_to_fp16, epsilon = var_6111_to_fp16, gamma = e_layers_14_norm_out_weight_to_fp16, x = input_619_cast_fp16)[name = string("input_621_cast_fp16")];
            tensor<int32, [1]> input_623_axes_0 = const()[name = string("input_623_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133969984)))];
            tensor<fp16, [1024]> e_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133972096)))];
            fp16 var_6118_to_fp16 = const()[name = string("op_6118_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_623_cast_fp16 = layer_norm(axes = input_623_axes_0, beta = e_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_6118_to_fp16, gamma = e_layers_15_norm_feed_forward1_weight_to_fp16, x = input_621_cast_fp16)[name = string("input_623_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_15_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133974208))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135022848))))[name = string("e_layers_15_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_15_feed_forward1_linear1_weight_to_fp16_lut, x = input_623_cast_fp16)[name = string("linear_136_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_627_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_627_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_15_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135285056))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136333696))))[name = string("e_layers_15_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_feed_forward1_linear2_weight_to_fp16_lut, x = input_627_cast_fp16)[name = string("linear_137_cast_fp16")];
            fp16 var_6135_to_fp16 = const()[name = string("op_6135_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6136_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_6135_to_fp16)[name = string("op_6136_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_631_cast_fp16 = add(x = input_621_cast_fp16, y = var_6136_cast_fp16)[name = string("input_631_cast_fp16")];
            tensor<int32, [1]> input_633_axes_0 = const()[name = string("input_633_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136595904)))];
            tensor<fp16, [1024]> e_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136598016)))];
            fp16 var_6140_to_fp16 = const()[name = string("op_6140_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_633_cast_fp16 = layer_norm(axes = input_633_axes_0, beta = e_layers_15_norm_self_att_bias_to_fp16, epsilon = var_6140_to_fp16, gamma = e_layers_15_norm_self_att_weight_to_fp16, x = input_631_cast_fp16)[name = string("input_633_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_15_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136600128))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136862336))))[name = string("e_layers_15_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_self_attn_q_proj_weight_to_fp16_lut, x = input_633_cast_fp16)[name = string("linear_138_cast_fp16")];
            tensor<int32, [4]> var_6153 = const()[name = string("op_6153"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6154_cast_fp16 = reshape(shape = var_6153, x = linear_138_cast_fp16)[name = string("op_6154_cast_fp16")];
            tensor<int32, [4]> q_31_perm_0 = const()[name = string("q_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_15_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136927936))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137190144))))[name = string("e_layers_15_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_self_attn_k_proj_weight_to_fp16_lut, x = input_633_cast_fp16)[name = string("linear_139_cast_fp16")];
            tensor<int32, [4]> var_6165 = const()[name = string("op_6165"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6166_cast_fp16 = reshape(shape = var_6165, x = linear_139_cast_fp16)[name = string("op_6166_cast_fp16")];
            tensor<int32, [4]> transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_15_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137255744))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137517952))))[name = string("e_layers_15_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_self_attn_v_proj_weight_to_fp16_lut, x = input_633_cast_fp16)[name = string("linear_140_cast_fp16")];
            tensor<int32, [4]> var_6177 = const()[name = string("op_6177"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6178_cast_fp16 = reshape(shape = var_6177, x = linear_140_cast_fp16)[name = string("op_6178_cast_fp16")];
            tensor<int32, [4]> v_31_perm_0 = const()[name = string("v_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_6202_to_fp16 = const()[name = string("op_6202_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137583552)))];
            tensor<fp16, [1, 8, 188, 128]> q_31_cast_fp16 = transpose(perm = q_31_perm_0, x = var_6154_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 8, 188, 128]> var_6204_cast_fp16 = add(x = q_31_cast_fp16, y = var_6202_to_fp16)[name = string("op_6204_cast_fp16")];
            bool input_635_transpose_x_0 = const()[name = string("input_635_transpose_x_0"), val = bool(false)];
            bool input_635_transpose_y_0 = const()[name = string("input_635_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_31_to_fp16 = const()[name = string("p_31_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137585664)))];
            tensor<fp16, [1, 8, 188, 375]> input_635_cast_fp16 = matmul(transpose_x = input_635_transpose_x_0, transpose_y = input_635_transpose_y_0, x = var_6204_cast_fp16, y = p_31_to_fp16)[name = string("input_635_cast_fp16")];
            tensor<int32, [8]> var_6211_pad_0 = const()[name = string("op_6211_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6211_mode_0 = const()[name = string("op_6211_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_6211_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = var_6211_mode_0, pad = var_6211_pad_0, x = input_635_cast_fp16)[name = string("op_6211_cast_fp16")];
            tensor<int32, [4]> var_6216 = const()[name = string("op_6216"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_6217_cast_fp16 = reshape(shape = var_6216, x = var_6211_cast_fp16)[name = string("op_6217_cast_fp16")];
            tensor<int32, [4]> var_6232_begin_0 = const()[name = string("op_6232_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_6232_end_0 = const()[name = string("op_6232_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_6232_end_mask_0 = const()[name = string("op_6232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_6232_cast_fp16 = slice_by_index(begin = var_6232_begin_0, end = var_6232_end_0, end_mask = var_6232_end_mask_0, x = var_6217_cast_fp16)[name = string("op_6232_cast_fp16")];
            tensor<int32, [4]> var_6237 = const()[name = string("op_6237"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_6238_cast_fp16 = reshape(shape = var_6237, x = var_6232_cast_fp16)[name = string("op_6238_cast_fp16")];
            tensor<int32, [4]> bd_31_begin_0 = const()[name = string("bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_31_end_0 = const()[name = string("bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_31_end_mask_0 = const()[name = string("bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_31_cast_fp16 = slice_by_index(begin = bd_31_begin_0, end = bd_31_end_0, end_mask = bd_31_end_mask_0, x = var_6238_cast_fp16)[name = string("bd_31_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_6249_to_fp16 = const()[name = string("op_6249_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138353728)))];
            tensor<fp16, [1, 8, 188, 128]> var_6251_cast_fp16 = add(x = q_31_cast_fp16, y = var_6249_to_fp16)[name = string("op_6251_cast_fp16")];
            bool ac_31_transpose_x_0 = const()[name = string("ac_31_transpose_x_0"), val = bool(false)];
            bool ac_31_transpose_y_0 = const()[name = string("ac_31_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_63_cast_fp16 = transpose(perm = transpose_63_perm_0, x = var_6166_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 8, 188, 188]> ac_31_cast_fp16 = matmul(transpose_x = ac_31_transpose_x_0, transpose_y = ac_31_transpose_y_0, x = var_6251_cast_fp16, y = transpose_63_cast_fp16)[name = string("ac_31_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6257_cast_fp16 = add(x = ac_31_cast_fp16, y = bd_31_cast_fp16)[name = string("op_6257_cast_fp16")];
            fp16 var_6258_to_fp16 = const()[name = string("op_6258_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6259_cast_fp16 = mul(x = var_6257_cast_fp16, y = var_6258_to_fp16)[name = string("op_6259_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6261_cast_fp16 = add(x = var_6259_cast_fp16, y = bias_15_cast_fp16)[name = string("op_6261_cast_fp16")];
            int32 var_6262 = const()[name = string("op_6262"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_31_cast_fp16 = softmax(axis = var_6262, x = var_6261_cast_fp16)[name = string("w_31_cast_fp16")];
            bool var_6265_transpose_x_0 = const()[name = string("op_6265_transpose_x_0"), val = bool(false)];
            bool var_6265_transpose_y_0 = const()[name = string("op_6265_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_31_cast_fp16 = transpose(perm = v_31_perm_0, x = var_6178_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 8, 188, 128]> var_6265_cast_fp16 = matmul(transpose_x = var_6265_transpose_x_0, transpose_y = var_6265_transpose_y_0, x = w_31_cast_fp16, y = v_31_cast_fp16)[name = string("op_6265_cast_fp16")];
            tensor<int32, [4]> var_6268_perm_0 = const()[name = string("op_6268_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6272 = const()[name = string("op_6272"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_6268_cast_fp16 = transpose(perm = var_6268_perm_0, x = var_6265_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 188, 1024]> input_637_cast_fp16 = reshape(shape = var_6272, x = var_6268_cast_fp16)[name = string("input_637_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_15_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138355840))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138618048))))[name = string("e_layers_15_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_self_attn_o_proj_weight_to_fp16_lut, x = input_637_cast_fp16)[name = string("linear_142_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_639_cast_fp16 = add(x = input_631_cast_fp16, y = linear_142_cast_fp16)[name = string("input_639_cast_fp16")];
            tensor<int32, [1]> x_91_axes_0 = const()[name = string("x_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_15_norm_conv_weight_to_fp16 = const()[name = string("e_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138683648)))];
            tensor<fp16, [1024]> e_layers_15_norm_conv_bias_to_fp16 = const()[name = string("e_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138685760)))];
            fp16 var_6280_to_fp16 = const()[name = string("op_6280_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_91_cast_fp16 = layer_norm(axes = x_91_axes_0, beta = e_layers_15_norm_conv_bias_to_fp16, epsilon = var_6280_to_fp16, gamma = e_layers_15_norm_conv_weight_to_fp16, x = input_639_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_15_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138687872))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139212224))))[name = string("e_layers_15_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_6294_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_15_conv_pointwise_conv1_weight_to_fp16_lut, x = x_91_cast_fp16)[name = string("op_6294_cast_fp16")];
            tensor<int32, [3]> input_641_perm_0 = const()[name = string("input_641_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_6297_split_num_splits_0 = const()[name = string("op_6297_split_num_splits_0"), val = int32(2)];
            int32 var_6297_split_axis_0 = const()[name = string("op_6297_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_641_cast_fp16 = transpose(perm = input_641_perm_0, x = var_6294_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 1024, 188]> var_6297_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_6297_split_cast_fp16_1 = split(axis = var_6297_split_axis_0, num_splits = var_6297_split_num_splits_0, x = input_641_cast_fp16)[name = string("op_6297_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6297_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6297_split_cast_fp16_1)[name = string("op_6297_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6297_cast_fp16 = mul(x = var_6297_split_cast_fp16_0, y = var_6297_split_1_sigmoid_cast_fp16)[name = string("op_6297_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_643_cast_fp16 = mul(x = var_6297_cast_fp16, y = var_2228_cast_fp16)[name = string("input_643_cast_fp16")];
            string input_645_pad_type_0 = const()[name = string("input_645_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_645_pad_0 = const()[name = string("input_645_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_645_groups_0 = const()[name = string("input_645_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_645_strides_0 = const()[name = string("input_645_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_645_dilations_0 = const()[name = string("input_645_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139343360)))];
            tensor<fp16, [1024]> const_72_to_fp16 = const()[name = string("const_72_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139361856)))];
            tensor<fp16, [1, 1024, 188]> input_647_cast_fp16 = conv(bias = const_72_to_fp16, dilations = input_645_dilations_0, groups = input_645_groups_0, pad = input_645_pad_0, pad_type = input_645_pad_type_0, strides = input_645_strides_0, weight = const_71_to_fp16, x = input_643_cast_fp16)[name = string("input_647_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_95_cast_fp16 = silu(x = input_647_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<int32, [3]> var_6337_perm_0 = const()[name = string("op_6337_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_15_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139363968))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139626176))))[name = string("e_layers_15_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_6337_cast_fp16 = transpose(perm = var_6337_perm_0, x = x_95_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 188, 1024]> var_6339_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_conv_pointwise_conv2_weight_to_fp16_lut, x = var_6337_cast_fp16)[name = string("op_6339_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_649_cast_fp16 = add(x = input_639_cast_fp16, y = var_6339_cast_fp16)[name = string("input_649_cast_fp16")];
            tensor<int32, [1]> input_651_axes_0 = const()[name = string("input_651_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139691776)))];
            tensor<fp16, [1024]> e_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139693888)))];
            fp16 var_6347_to_fp16 = const()[name = string("op_6347_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_651_cast_fp16 = layer_norm(axes = input_651_axes_0, beta = e_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_6347_to_fp16, gamma = e_layers_15_norm_feed_forward2_weight_to_fp16, x = input_649_cast_fp16)[name = string("input_651_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_15_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139696000))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140744640))))[name = string("e_layers_15_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_15_feed_forward2_linear1_weight_to_fp16_lut, x = input_651_cast_fp16)[name = string("linear_143_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_655_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_655_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_15_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141006848))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142055488))))[name = string("e_layers_15_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_15_feed_forward2_linear2_weight_to_fp16_lut, x = input_655_cast_fp16)[name = string("linear_144_cast_fp16")];
            fp16 var_6364_to_fp16 = const()[name = string("op_6364_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6365_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_6364_to_fp16)[name = string("op_6365_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_659_cast_fp16 = add(x = input_649_cast_fp16, y = var_6365_cast_fp16)[name = string("input_659_cast_fp16")];
            tensor<int32, [1]> input_661_axes_0 = const()[name = string("input_661_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_15_norm_out_weight_to_fp16 = const()[name = string("e_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142317696)))];
            tensor<fp16, [1024]> e_layers_15_norm_out_bias_to_fp16 = const()[name = string("e_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142319808)))];
            fp16 var_6369_to_fp16 = const()[name = string("op_6369_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_661_cast_fp16 = layer_norm(axes = input_661_axes_0, beta = e_layers_15_norm_out_bias_to_fp16, epsilon = var_6369_to_fp16, gamma = e_layers_15_norm_out_weight_to_fp16, x = input_659_cast_fp16)[name = string("input_661_cast_fp16")];
            tensor<int32, [1]> input_663_axes_0 = const()[name = string("input_663_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142321920)))];
            tensor<fp16, [1024]> e_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142324032)))];
            fp16 var_6376_to_fp16 = const()[name = string("op_6376_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_663_cast_fp16 = layer_norm(axes = input_663_axes_0, beta = e_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_6376_to_fp16, gamma = e_layers_16_norm_feed_forward1_weight_to_fp16, x = input_661_cast_fp16)[name = string("input_663_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_16_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142326144))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143374784))))[name = string("e_layers_16_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_16_feed_forward1_linear1_weight_to_fp16_lut, x = input_663_cast_fp16)[name = string("linear_145_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_667_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_667_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_16_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143636992))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144685632))))[name = string("e_layers_16_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_feed_forward1_linear2_weight_to_fp16_lut, x = input_667_cast_fp16)[name = string("linear_146_cast_fp16")];
            fp16 var_6393_to_fp16 = const()[name = string("op_6393_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6394_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_6393_to_fp16)[name = string("op_6394_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_671_cast_fp16 = add(x = input_661_cast_fp16, y = var_6394_cast_fp16)[name = string("input_671_cast_fp16")];
            tensor<int32, [1]> input_673_axes_0 = const()[name = string("input_673_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144947840)))];
            tensor<fp16, [1024]> e_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144949952)))];
            fp16 var_6398_to_fp16 = const()[name = string("op_6398_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_673_cast_fp16 = layer_norm(axes = input_673_axes_0, beta = e_layers_16_norm_self_att_bias_to_fp16, epsilon = var_6398_to_fp16, gamma = e_layers_16_norm_self_att_weight_to_fp16, x = input_671_cast_fp16)[name = string("input_673_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_16_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144952064))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145214272))))[name = string("e_layers_16_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_self_attn_q_proj_weight_to_fp16_lut, x = input_673_cast_fp16)[name = string("linear_147_cast_fp16")];
            tensor<int32, [4]> var_6411 = const()[name = string("op_6411"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6412_cast_fp16 = reshape(shape = var_6411, x = linear_147_cast_fp16)[name = string("op_6412_cast_fp16")];
            tensor<int32, [4]> q_33_perm_0 = const()[name = string("q_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_16_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145279872))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145542080))))[name = string("e_layers_16_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_self_attn_k_proj_weight_to_fp16_lut, x = input_673_cast_fp16)[name = string("linear_148_cast_fp16")];
            tensor<int32, [4]> var_6423 = const()[name = string("op_6423"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6424_cast_fp16 = reshape(shape = var_6423, x = linear_148_cast_fp16)[name = string("op_6424_cast_fp16")];
            tensor<int32, [4]> transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_16_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145607680))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145869888))))[name = string("e_layers_16_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_self_attn_v_proj_weight_to_fp16_lut, x = input_673_cast_fp16)[name = string("linear_149_cast_fp16")];
            tensor<int32, [4]> var_6435 = const()[name = string("op_6435"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6436_cast_fp16 = reshape(shape = var_6435, x = linear_149_cast_fp16)[name = string("op_6436_cast_fp16")];
            tensor<int32, [4]> v_33_perm_0 = const()[name = string("v_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_6460_to_fp16 = const()[name = string("op_6460_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145935488)))];
            tensor<fp16, [1, 8, 188, 128]> q_33_cast_fp16 = transpose(perm = q_33_perm_0, x = var_6412_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 8, 188, 128]> var_6462_cast_fp16 = add(x = q_33_cast_fp16, y = var_6460_to_fp16)[name = string("op_6462_cast_fp16")];
            bool input_675_transpose_x_0 = const()[name = string("input_675_transpose_x_0"), val = bool(false)];
            bool input_675_transpose_y_0 = const()[name = string("input_675_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_33_to_fp16 = const()[name = string("p_33_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145937600)))];
            tensor<fp16, [1, 8, 188, 375]> input_675_cast_fp16 = matmul(transpose_x = input_675_transpose_x_0, transpose_y = input_675_transpose_y_0, x = var_6462_cast_fp16, y = p_33_to_fp16)[name = string("input_675_cast_fp16")];
            tensor<int32, [8]> var_6469_pad_0 = const()[name = string("op_6469_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6469_mode_0 = const()[name = string("op_6469_mode_0"), val = string("constant")];
            fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_6469_cast_fp16 = pad(constant_val = const_33_to_fp16, mode = var_6469_mode_0, pad = var_6469_pad_0, x = input_675_cast_fp16)[name = string("op_6469_cast_fp16")];
            tensor<int32, [4]> var_6474 = const()[name = string("op_6474"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_6475_cast_fp16 = reshape(shape = var_6474, x = var_6469_cast_fp16)[name = string("op_6475_cast_fp16")];
            tensor<int32, [4]> var_6490_begin_0 = const()[name = string("op_6490_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_6490_end_0 = const()[name = string("op_6490_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_6490_end_mask_0 = const()[name = string("op_6490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_6490_cast_fp16 = slice_by_index(begin = var_6490_begin_0, end = var_6490_end_0, end_mask = var_6490_end_mask_0, x = var_6475_cast_fp16)[name = string("op_6490_cast_fp16")];
            tensor<int32, [4]> var_6495 = const()[name = string("op_6495"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_6496_cast_fp16 = reshape(shape = var_6495, x = var_6490_cast_fp16)[name = string("op_6496_cast_fp16")];
            tensor<int32, [4]> bd_33_begin_0 = const()[name = string("bd_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_33_end_0 = const()[name = string("bd_33_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_33_end_mask_0 = const()[name = string("bd_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_33_cast_fp16 = slice_by_index(begin = bd_33_begin_0, end = bd_33_end_0, end_mask = bd_33_end_mask_0, x = var_6496_cast_fp16)[name = string("bd_33_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_6507_to_fp16 = const()[name = string("op_6507_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146705664)))];
            tensor<fp16, [1, 8, 188, 128]> var_6509_cast_fp16 = add(x = q_33_cast_fp16, y = var_6507_to_fp16)[name = string("op_6509_cast_fp16")];
            bool ac_33_transpose_x_0 = const()[name = string("ac_33_transpose_x_0"), val = bool(false)];
            bool ac_33_transpose_y_0 = const()[name = string("ac_33_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_64_cast_fp16 = transpose(perm = transpose_64_perm_0, x = var_6424_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 8, 188, 188]> ac_33_cast_fp16 = matmul(transpose_x = ac_33_transpose_x_0, transpose_y = ac_33_transpose_y_0, x = var_6509_cast_fp16, y = transpose_64_cast_fp16)[name = string("ac_33_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6515_cast_fp16 = add(x = ac_33_cast_fp16, y = bd_33_cast_fp16)[name = string("op_6515_cast_fp16")];
            fp16 var_6516_to_fp16 = const()[name = string("op_6516_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6517_cast_fp16 = mul(x = var_6515_cast_fp16, y = var_6516_to_fp16)[name = string("op_6517_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6519_cast_fp16 = add(x = var_6517_cast_fp16, y = bias_15_cast_fp16)[name = string("op_6519_cast_fp16")];
            int32 var_6520 = const()[name = string("op_6520"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_33_cast_fp16 = softmax(axis = var_6520, x = var_6519_cast_fp16)[name = string("w_33_cast_fp16")];
            bool var_6523_transpose_x_0 = const()[name = string("op_6523_transpose_x_0"), val = bool(false)];
            bool var_6523_transpose_y_0 = const()[name = string("op_6523_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_33_cast_fp16 = transpose(perm = v_33_perm_0, x = var_6436_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 8, 188, 128]> var_6523_cast_fp16 = matmul(transpose_x = var_6523_transpose_x_0, transpose_y = var_6523_transpose_y_0, x = w_33_cast_fp16, y = v_33_cast_fp16)[name = string("op_6523_cast_fp16")];
            tensor<int32, [4]> var_6526_perm_0 = const()[name = string("op_6526_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6530 = const()[name = string("op_6530"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_6526_cast_fp16 = transpose(perm = var_6526_perm_0, x = var_6523_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 188, 1024]> input_677_cast_fp16 = reshape(shape = var_6530, x = var_6526_cast_fp16)[name = string("input_677_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_16_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146707776))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146969984))))[name = string("e_layers_16_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_self_attn_o_proj_weight_to_fp16_lut, x = input_677_cast_fp16)[name = string("linear_151_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_679_cast_fp16 = add(x = input_671_cast_fp16, y = linear_151_cast_fp16)[name = string("input_679_cast_fp16")];
            tensor<int32, [1]> x_97_axes_0 = const()[name = string("x_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_16_norm_conv_weight_to_fp16 = const()[name = string("e_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147035584)))];
            tensor<fp16, [1024]> e_layers_16_norm_conv_bias_to_fp16 = const()[name = string("e_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147037696)))];
            fp16 var_6538_to_fp16 = const()[name = string("op_6538_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_97_cast_fp16 = layer_norm(axes = x_97_axes_0, beta = e_layers_16_norm_conv_bias_to_fp16, epsilon = var_6538_to_fp16, gamma = e_layers_16_norm_conv_weight_to_fp16, x = input_679_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_16_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147039808))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147564160))))[name = string("e_layers_16_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_6552_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_16_conv_pointwise_conv1_weight_to_fp16_lut, x = x_97_cast_fp16)[name = string("op_6552_cast_fp16")];
            tensor<int32, [3]> input_681_perm_0 = const()[name = string("input_681_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_6555_split_num_splits_0 = const()[name = string("op_6555_split_num_splits_0"), val = int32(2)];
            int32 var_6555_split_axis_0 = const()[name = string("op_6555_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_681_cast_fp16 = transpose(perm = input_681_perm_0, x = var_6552_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 1024, 188]> var_6555_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_6555_split_cast_fp16_1 = split(axis = var_6555_split_axis_0, num_splits = var_6555_split_num_splits_0, x = input_681_cast_fp16)[name = string("op_6555_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6555_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6555_split_cast_fp16_1)[name = string("op_6555_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6555_cast_fp16 = mul(x = var_6555_split_cast_fp16_0, y = var_6555_split_1_sigmoid_cast_fp16)[name = string("op_6555_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_683_cast_fp16 = mul(x = var_6555_cast_fp16, y = var_2228_cast_fp16)[name = string("input_683_cast_fp16")];
            string input_685_pad_type_0 = const()[name = string("input_685_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_685_pad_0 = const()[name = string("input_685_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_685_groups_0 = const()[name = string("input_685_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_685_strides_0 = const()[name = string("input_685_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_685_dilations_0 = const()[name = string("input_685_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_73_to_fp16 = const()[name = string("const_73_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147695296)))];
            tensor<fp16, [1024]> const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147713792)))];
            tensor<fp16, [1, 1024, 188]> input_687_cast_fp16 = conv(bias = const_74_to_fp16, dilations = input_685_dilations_0, groups = input_685_groups_0, pad = input_685_pad_0, pad_type = input_685_pad_type_0, strides = input_685_strides_0, weight = const_73_to_fp16, x = input_683_cast_fp16)[name = string("input_687_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_101_cast_fp16 = silu(x = input_687_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [3]> var_6595_perm_0 = const()[name = string("op_6595_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_16_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147715904))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147978112))))[name = string("e_layers_16_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_6595_cast_fp16 = transpose(perm = var_6595_perm_0, x = x_101_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 188, 1024]> var_6597_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_conv_pointwise_conv2_weight_to_fp16_lut, x = var_6595_cast_fp16)[name = string("op_6597_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_689_cast_fp16 = add(x = input_679_cast_fp16, y = var_6597_cast_fp16)[name = string("input_689_cast_fp16")];
            tensor<int32, [1]> input_691_axes_0 = const()[name = string("input_691_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148043712)))];
            tensor<fp16, [1024]> e_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148045824)))];
            fp16 var_6605_to_fp16 = const()[name = string("op_6605_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_691_cast_fp16 = layer_norm(axes = input_691_axes_0, beta = e_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_6605_to_fp16, gamma = e_layers_16_norm_feed_forward2_weight_to_fp16, x = input_689_cast_fp16)[name = string("input_691_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_16_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148047936))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149096576))))[name = string("e_layers_16_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_16_feed_forward2_linear1_weight_to_fp16_lut, x = input_691_cast_fp16)[name = string("linear_152_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_695_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_695_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_16_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149358784))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150407424))))[name = string("e_layers_16_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_16_feed_forward2_linear2_weight_to_fp16_lut, x = input_695_cast_fp16)[name = string("linear_153_cast_fp16")];
            fp16 var_6622_to_fp16 = const()[name = string("op_6622_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6623_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_6622_to_fp16)[name = string("op_6623_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_699_cast_fp16 = add(x = input_689_cast_fp16, y = var_6623_cast_fp16)[name = string("input_699_cast_fp16")];
            tensor<int32, [1]> input_701_axes_0 = const()[name = string("input_701_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_16_norm_out_weight_to_fp16 = const()[name = string("e_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150669632)))];
            tensor<fp16, [1024]> e_layers_16_norm_out_bias_to_fp16 = const()[name = string("e_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150671744)))];
            fp16 var_6627_to_fp16 = const()[name = string("op_6627_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_701_cast_fp16 = layer_norm(axes = input_701_axes_0, beta = e_layers_16_norm_out_bias_to_fp16, epsilon = var_6627_to_fp16, gamma = e_layers_16_norm_out_weight_to_fp16, x = input_699_cast_fp16)[name = string("input_701_cast_fp16")];
            tensor<int32, [1]> input_703_axes_0 = const()[name = string("input_703_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150673856)))];
            tensor<fp16, [1024]> e_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150675968)))];
            fp16 var_6634_to_fp16 = const()[name = string("op_6634_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_703_cast_fp16 = layer_norm(axes = input_703_axes_0, beta = e_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_6634_to_fp16, gamma = e_layers_17_norm_feed_forward1_weight_to_fp16, x = input_701_cast_fp16)[name = string("input_703_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_17_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150678080))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151726720))))[name = string("e_layers_17_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_17_feed_forward1_linear1_weight_to_fp16_lut, x = input_703_cast_fp16)[name = string("linear_154_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_707_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_707_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_17_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151988928))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153037568))))[name = string("e_layers_17_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_feed_forward1_linear2_weight_to_fp16_lut, x = input_707_cast_fp16)[name = string("linear_155_cast_fp16")];
            fp16 var_6651_to_fp16 = const()[name = string("op_6651_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6652_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_6651_to_fp16)[name = string("op_6652_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_711_cast_fp16 = add(x = input_701_cast_fp16, y = var_6652_cast_fp16)[name = string("input_711_cast_fp16")];
            tensor<int32, [1]> input_713_axes_0 = const()[name = string("input_713_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153299776)))];
            tensor<fp16, [1024]> e_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153301888)))];
            fp16 var_6656_to_fp16 = const()[name = string("op_6656_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_713_cast_fp16 = layer_norm(axes = input_713_axes_0, beta = e_layers_17_norm_self_att_bias_to_fp16, epsilon = var_6656_to_fp16, gamma = e_layers_17_norm_self_att_weight_to_fp16, x = input_711_cast_fp16)[name = string("input_713_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_17_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153304000))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153566208))))[name = string("e_layers_17_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_self_attn_q_proj_weight_to_fp16_lut, x = input_713_cast_fp16)[name = string("linear_156_cast_fp16")];
            tensor<int32, [4]> var_6669 = const()[name = string("op_6669"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6670_cast_fp16 = reshape(shape = var_6669, x = linear_156_cast_fp16)[name = string("op_6670_cast_fp16")];
            tensor<int32, [4]> q_35_perm_0 = const()[name = string("q_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_17_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153631808))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153894016))))[name = string("e_layers_17_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_self_attn_k_proj_weight_to_fp16_lut, x = input_713_cast_fp16)[name = string("linear_157_cast_fp16")];
            tensor<int32, [4]> var_6681 = const()[name = string("op_6681"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6682_cast_fp16 = reshape(shape = var_6681, x = linear_157_cast_fp16)[name = string("op_6682_cast_fp16")];
            tensor<int32, [4]> transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_17_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153959616))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154221824))))[name = string("e_layers_17_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_self_attn_v_proj_weight_to_fp16_lut, x = input_713_cast_fp16)[name = string("linear_158_cast_fp16")];
            tensor<int32, [4]> var_6693 = const()[name = string("op_6693"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6694_cast_fp16 = reshape(shape = var_6693, x = linear_158_cast_fp16)[name = string("op_6694_cast_fp16")];
            tensor<int32, [4]> v_35_perm_0 = const()[name = string("v_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_6718_to_fp16 = const()[name = string("op_6718_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154287424)))];
            tensor<fp16, [1, 8, 188, 128]> q_35_cast_fp16 = transpose(perm = q_35_perm_0, x = var_6670_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 8, 188, 128]> var_6720_cast_fp16 = add(x = q_35_cast_fp16, y = var_6718_to_fp16)[name = string("op_6720_cast_fp16")];
            bool input_715_transpose_x_0 = const()[name = string("input_715_transpose_x_0"), val = bool(false)];
            bool input_715_transpose_y_0 = const()[name = string("input_715_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_35_to_fp16 = const()[name = string("p_35_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154289536)))];
            tensor<fp16, [1, 8, 188, 375]> input_715_cast_fp16 = matmul(transpose_x = input_715_transpose_x_0, transpose_y = input_715_transpose_y_0, x = var_6720_cast_fp16, y = p_35_to_fp16)[name = string("input_715_cast_fp16")];
            tensor<int32, [8]> var_6727_pad_0 = const()[name = string("op_6727_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6727_mode_0 = const()[name = string("op_6727_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_6727_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = var_6727_mode_0, pad = var_6727_pad_0, x = input_715_cast_fp16)[name = string("op_6727_cast_fp16")];
            tensor<int32, [4]> var_6732 = const()[name = string("op_6732"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_6733_cast_fp16 = reshape(shape = var_6732, x = var_6727_cast_fp16)[name = string("op_6733_cast_fp16")];
            tensor<int32, [4]> var_6748_begin_0 = const()[name = string("op_6748_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_6748_end_0 = const()[name = string("op_6748_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_6748_end_mask_0 = const()[name = string("op_6748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_6748_cast_fp16 = slice_by_index(begin = var_6748_begin_0, end = var_6748_end_0, end_mask = var_6748_end_mask_0, x = var_6733_cast_fp16)[name = string("op_6748_cast_fp16")];
            tensor<int32, [4]> var_6753 = const()[name = string("op_6753"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_6754_cast_fp16 = reshape(shape = var_6753, x = var_6748_cast_fp16)[name = string("op_6754_cast_fp16")];
            tensor<int32, [4]> bd_35_begin_0 = const()[name = string("bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_35_end_0 = const()[name = string("bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_35_end_mask_0 = const()[name = string("bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_35_cast_fp16 = slice_by_index(begin = bd_35_begin_0, end = bd_35_end_0, end_mask = bd_35_end_mask_0, x = var_6754_cast_fp16)[name = string("bd_35_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_6765_to_fp16 = const()[name = string("op_6765_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155057600)))];
            tensor<fp16, [1, 8, 188, 128]> var_6767_cast_fp16 = add(x = q_35_cast_fp16, y = var_6765_to_fp16)[name = string("op_6767_cast_fp16")];
            bool ac_35_transpose_x_0 = const()[name = string("ac_35_transpose_x_0"), val = bool(false)];
            bool ac_35_transpose_y_0 = const()[name = string("ac_35_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_65_cast_fp16 = transpose(perm = transpose_65_perm_0, x = var_6682_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 8, 188, 188]> ac_35_cast_fp16 = matmul(transpose_x = ac_35_transpose_x_0, transpose_y = ac_35_transpose_y_0, x = var_6767_cast_fp16, y = transpose_65_cast_fp16)[name = string("ac_35_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6773_cast_fp16 = add(x = ac_35_cast_fp16, y = bd_35_cast_fp16)[name = string("op_6773_cast_fp16")];
            fp16 var_6774_to_fp16 = const()[name = string("op_6774_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_6775_cast_fp16 = mul(x = var_6773_cast_fp16, y = var_6774_to_fp16)[name = string("op_6775_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_6777_cast_fp16 = add(x = var_6775_cast_fp16, y = bias_15_cast_fp16)[name = string("op_6777_cast_fp16")];
            int32 var_6778 = const()[name = string("op_6778"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_35_cast_fp16 = softmax(axis = var_6778, x = var_6777_cast_fp16)[name = string("w_35_cast_fp16")];
            bool var_6781_transpose_x_0 = const()[name = string("op_6781_transpose_x_0"), val = bool(false)];
            bool var_6781_transpose_y_0 = const()[name = string("op_6781_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_35_cast_fp16 = transpose(perm = v_35_perm_0, x = var_6694_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 8, 188, 128]> var_6781_cast_fp16 = matmul(transpose_x = var_6781_transpose_x_0, transpose_y = var_6781_transpose_y_0, x = w_35_cast_fp16, y = v_35_cast_fp16)[name = string("op_6781_cast_fp16")];
            tensor<int32, [4]> var_6784_perm_0 = const()[name = string("op_6784_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6788 = const()[name = string("op_6788"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_6784_cast_fp16 = transpose(perm = var_6784_perm_0, x = var_6781_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 188, 1024]> input_717_cast_fp16 = reshape(shape = var_6788, x = var_6784_cast_fp16)[name = string("input_717_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_17_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155059712))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155321920))))[name = string("e_layers_17_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_self_attn_o_proj_weight_to_fp16_lut, x = input_717_cast_fp16)[name = string("linear_160_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_719_cast_fp16 = add(x = input_711_cast_fp16, y = linear_160_cast_fp16)[name = string("input_719_cast_fp16")];
            tensor<int32, [1]> x_103_axes_0 = const()[name = string("x_103_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_17_norm_conv_weight_to_fp16 = const()[name = string("e_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155387520)))];
            tensor<fp16, [1024]> e_layers_17_norm_conv_bias_to_fp16 = const()[name = string("e_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155389632)))];
            fp16 var_6796_to_fp16 = const()[name = string("op_6796_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_103_cast_fp16 = layer_norm(axes = x_103_axes_0, beta = e_layers_17_norm_conv_bias_to_fp16, epsilon = var_6796_to_fp16, gamma = e_layers_17_norm_conv_weight_to_fp16, x = input_719_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_17_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155391744))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155916096))))[name = string("e_layers_17_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_6810_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_17_conv_pointwise_conv1_weight_to_fp16_lut, x = x_103_cast_fp16)[name = string("op_6810_cast_fp16")];
            tensor<int32, [3]> input_721_perm_0 = const()[name = string("input_721_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_6813_split_num_splits_0 = const()[name = string("op_6813_split_num_splits_0"), val = int32(2)];
            int32 var_6813_split_axis_0 = const()[name = string("op_6813_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_721_cast_fp16 = transpose(perm = input_721_perm_0, x = var_6810_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 1024, 188]> var_6813_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_6813_split_cast_fp16_1 = split(axis = var_6813_split_axis_0, num_splits = var_6813_split_num_splits_0, x = input_721_cast_fp16)[name = string("op_6813_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6813_split_1_sigmoid_cast_fp16 = sigmoid(x = var_6813_split_cast_fp16_1)[name = string("op_6813_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_6813_cast_fp16 = mul(x = var_6813_split_cast_fp16_0, y = var_6813_split_1_sigmoid_cast_fp16)[name = string("op_6813_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_723_cast_fp16 = mul(x = var_6813_cast_fp16, y = var_2228_cast_fp16)[name = string("input_723_cast_fp16")];
            string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_75_to_fp16 = const()[name = string("const_75_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156047232)))];
            tensor<fp16, [1024]> const_76_to_fp16 = const()[name = string("const_76_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156065728)))];
            tensor<fp16, [1, 1024, 188]> input_727_cast_fp16 = conv(bias = const_76_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_75_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_107_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [3]> var_6853_perm_0 = const()[name = string("op_6853_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_17_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156067840))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156330048))))[name = string("e_layers_17_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_6853_cast_fp16 = transpose(perm = var_6853_perm_0, x = x_107_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 188, 1024]> var_6855_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_conv_pointwise_conv2_weight_to_fp16_lut, x = var_6853_cast_fp16)[name = string("op_6855_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_729_cast_fp16 = add(x = input_719_cast_fp16, y = var_6855_cast_fp16)[name = string("input_729_cast_fp16")];
            tensor<int32, [1]> input_731_axes_0 = const()[name = string("input_731_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156395648)))];
            tensor<fp16, [1024]> e_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156397760)))];
            fp16 var_6863_to_fp16 = const()[name = string("op_6863_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_731_cast_fp16 = layer_norm(axes = input_731_axes_0, beta = e_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_6863_to_fp16, gamma = e_layers_17_norm_feed_forward2_weight_to_fp16, x = input_729_cast_fp16)[name = string("input_731_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_17_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156399872))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157448512))))[name = string("e_layers_17_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_17_feed_forward2_linear1_weight_to_fp16_lut, x = input_731_cast_fp16)[name = string("linear_161_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_735_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_735_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_17_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157710720))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158759360))))[name = string("e_layers_17_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_17_feed_forward2_linear2_weight_to_fp16_lut, x = input_735_cast_fp16)[name = string("linear_162_cast_fp16")];
            fp16 var_6880_to_fp16 = const()[name = string("op_6880_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6881_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_6880_to_fp16)[name = string("op_6881_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_739_cast_fp16 = add(x = input_729_cast_fp16, y = var_6881_cast_fp16)[name = string("input_739_cast_fp16")];
            tensor<int32, [1]> input_741_axes_0 = const()[name = string("input_741_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_17_norm_out_weight_to_fp16 = const()[name = string("e_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159021568)))];
            tensor<fp16, [1024]> e_layers_17_norm_out_bias_to_fp16 = const()[name = string("e_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159023680)))];
            fp16 var_6885_to_fp16 = const()[name = string("op_6885_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_741_cast_fp16 = layer_norm(axes = input_741_axes_0, beta = e_layers_17_norm_out_bias_to_fp16, epsilon = var_6885_to_fp16, gamma = e_layers_17_norm_out_weight_to_fp16, x = input_739_cast_fp16)[name = string("input_741_cast_fp16")];
            tensor<int32, [1]> input_743_axes_0 = const()[name = string("input_743_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159025792)))];
            tensor<fp16, [1024]> e_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159027904)))];
            fp16 var_6892_to_fp16 = const()[name = string("op_6892_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_743_cast_fp16 = layer_norm(axes = input_743_axes_0, beta = e_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_6892_to_fp16, gamma = e_layers_18_norm_feed_forward1_weight_to_fp16, x = input_741_cast_fp16)[name = string("input_743_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_18_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159030016))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160078656))))[name = string("e_layers_18_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_18_feed_forward1_linear1_weight_to_fp16_lut, x = input_743_cast_fp16)[name = string("linear_163_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_747_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_747_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_18_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160340864))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161389504))))[name = string("e_layers_18_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_feed_forward1_linear2_weight_to_fp16_lut, x = input_747_cast_fp16)[name = string("linear_164_cast_fp16")];
            fp16 var_6909_to_fp16 = const()[name = string("op_6909_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_6910_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_6909_to_fp16)[name = string("op_6910_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_751_cast_fp16 = add(x = input_741_cast_fp16, y = var_6910_cast_fp16)[name = string("input_751_cast_fp16")];
            tensor<int32, [1]> input_753_axes_0 = const()[name = string("input_753_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161651712)))];
            tensor<fp16, [1024]> e_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161653824)))];
            fp16 var_6914_to_fp16 = const()[name = string("op_6914_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_753_cast_fp16 = layer_norm(axes = input_753_axes_0, beta = e_layers_18_norm_self_att_bias_to_fp16, epsilon = var_6914_to_fp16, gamma = e_layers_18_norm_self_att_weight_to_fp16, x = input_751_cast_fp16)[name = string("input_753_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_18_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161655936))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161918144))))[name = string("e_layers_18_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_self_attn_q_proj_weight_to_fp16_lut, x = input_753_cast_fp16)[name = string("linear_165_cast_fp16")];
            tensor<int32, [4]> var_6927 = const()[name = string("op_6927"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6928_cast_fp16 = reshape(shape = var_6927, x = linear_165_cast_fp16)[name = string("op_6928_cast_fp16")];
            tensor<int32, [4]> q_37_perm_0 = const()[name = string("q_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_18_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161983744))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162245952))))[name = string("e_layers_18_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_self_attn_k_proj_weight_to_fp16_lut, x = input_753_cast_fp16)[name = string("linear_166_cast_fp16")];
            tensor<int32, [4]> var_6939 = const()[name = string("op_6939"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6940_cast_fp16 = reshape(shape = var_6939, x = linear_166_cast_fp16)[name = string("op_6940_cast_fp16")];
            tensor<int32, [4]> transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_18_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162311552))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162573760))))[name = string("e_layers_18_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_self_attn_v_proj_weight_to_fp16_lut, x = input_753_cast_fp16)[name = string("linear_167_cast_fp16")];
            tensor<int32, [4]> var_6951 = const()[name = string("op_6951"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_6952_cast_fp16 = reshape(shape = var_6951, x = linear_167_cast_fp16)[name = string("op_6952_cast_fp16")];
            tensor<int32, [4]> v_37_perm_0 = const()[name = string("v_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_6976_to_fp16 = const()[name = string("op_6976_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162639360)))];
            tensor<fp16, [1, 8, 188, 128]> q_37_cast_fp16 = transpose(perm = q_37_perm_0, x = var_6928_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 8, 188, 128]> var_6978_cast_fp16 = add(x = q_37_cast_fp16, y = var_6976_to_fp16)[name = string("op_6978_cast_fp16")];
            bool input_755_transpose_x_0 = const()[name = string("input_755_transpose_x_0"), val = bool(false)];
            bool input_755_transpose_y_0 = const()[name = string("input_755_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_37_to_fp16 = const()[name = string("p_37_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162641472)))];
            tensor<fp16, [1, 8, 188, 375]> input_755_cast_fp16 = matmul(transpose_x = input_755_transpose_x_0, transpose_y = input_755_transpose_y_0, x = var_6978_cast_fp16, y = p_37_to_fp16)[name = string("input_755_cast_fp16")];
            tensor<int32, [8]> var_6985_pad_0 = const()[name = string("op_6985_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_6985_mode_0 = const()[name = string("op_6985_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_6985_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = var_6985_mode_0, pad = var_6985_pad_0, x = input_755_cast_fp16)[name = string("op_6985_cast_fp16")];
            tensor<int32, [4]> var_6990 = const()[name = string("op_6990"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_6991_cast_fp16 = reshape(shape = var_6990, x = var_6985_cast_fp16)[name = string("op_6991_cast_fp16")];
            tensor<int32, [4]> var_7006_begin_0 = const()[name = string("op_7006_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_7006_end_0 = const()[name = string("op_7006_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_7006_end_mask_0 = const()[name = string("op_7006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_7006_cast_fp16 = slice_by_index(begin = var_7006_begin_0, end = var_7006_end_0, end_mask = var_7006_end_mask_0, x = var_6991_cast_fp16)[name = string("op_7006_cast_fp16")];
            tensor<int32, [4]> var_7011 = const()[name = string("op_7011"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_7012_cast_fp16 = reshape(shape = var_7011, x = var_7006_cast_fp16)[name = string("op_7012_cast_fp16")];
            tensor<int32, [4]> bd_37_begin_0 = const()[name = string("bd_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_37_end_0 = const()[name = string("bd_37_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_37_end_mask_0 = const()[name = string("bd_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_37_cast_fp16 = slice_by_index(begin = bd_37_begin_0, end = bd_37_end_0, end_mask = bd_37_end_mask_0, x = var_7012_cast_fp16)[name = string("bd_37_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_7023_to_fp16 = const()[name = string("op_7023_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163409536)))];
            tensor<fp16, [1, 8, 188, 128]> var_7025_cast_fp16 = add(x = q_37_cast_fp16, y = var_7023_to_fp16)[name = string("op_7025_cast_fp16")];
            bool ac_37_transpose_x_0 = const()[name = string("ac_37_transpose_x_0"), val = bool(false)];
            bool ac_37_transpose_y_0 = const()[name = string("ac_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_66_cast_fp16 = transpose(perm = transpose_66_perm_0, x = var_6940_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 8, 188, 188]> ac_37_cast_fp16 = matmul(transpose_x = ac_37_transpose_x_0, transpose_y = ac_37_transpose_y_0, x = var_7025_cast_fp16, y = transpose_66_cast_fp16)[name = string("ac_37_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7031_cast_fp16 = add(x = ac_37_cast_fp16, y = bd_37_cast_fp16)[name = string("op_7031_cast_fp16")];
            fp16 var_7032_to_fp16 = const()[name = string("op_7032_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_7033_cast_fp16 = mul(x = var_7031_cast_fp16, y = var_7032_to_fp16)[name = string("op_7033_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7035_cast_fp16 = add(x = var_7033_cast_fp16, y = bias_15_cast_fp16)[name = string("op_7035_cast_fp16")];
            int32 var_7036 = const()[name = string("op_7036"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_37_cast_fp16 = softmax(axis = var_7036, x = var_7035_cast_fp16)[name = string("w_37_cast_fp16")];
            bool var_7039_transpose_x_0 = const()[name = string("op_7039_transpose_x_0"), val = bool(false)];
            bool var_7039_transpose_y_0 = const()[name = string("op_7039_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_37_cast_fp16 = transpose(perm = v_37_perm_0, x = var_6952_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 8, 188, 128]> var_7039_cast_fp16 = matmul(transpose_x = var_7039_transpose_x_0, transpose_y = var_7039_transpose_y_0, x = w_37_cast_fp16, y = v_37_cast_fp16)[name = string("op_7039_cast_fp16")];
            tensor<int32, [4]> var_7042_perm_0 = const()[name = string("op_7042_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7046 = const()[name = string("op_7046"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_7042_cast_fp16 = transpose(perm = var_7042_perm_0, x = var_7039_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 188, 1024]> input_757_cast_fp16 = reshape(shape = var_7046, x = var_7042_cast_fp16)[name = string("input_757_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_18_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163411648))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163673856))))[name = string("e_layers_18_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_self_attn_o_proj_weight_to_fp16_lut, x = input_757_cast_fp16)[name = string("linear_169_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_759_cast_fp16 = add(x = input_751_cast_fp16, y = linear_169_cast_fp16)[name = string("input_759_cast_fp16")];
            tensor<int32, [1]> x_109_axes_0 = const()[name = string("x_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_18_norm_conv_weight_to_fp16 = const()[name = string("e_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163739456)))];
            tensor<fp16, [1024]> e_layers_18_norm_conv_bias_to_fp16 = const()[name = string("e_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163741568)))];
            fp16 var_7054_to_fp16 = const()[name = string("op_7054_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_109_cast_fp16 = layer_norm(axes = x_109_axes_0, beta = e_layers_18_norm_conv_bias_to_fp16, epsilon = var_7054_to_fp16, gamma = e_layers_18_norm_conv_weight_to_fp16, x = input_759_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_18_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163743680))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164268032))))[name = string("e_layers_18_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_7068_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_18_conv_pointwise_conv1_weight_to_fp16_lut, x = x_109_cast_fp16)[name = string("op_7068_cast_fp16")];
            tensor<int32, [3]> input_761_perm_0 = const()[name = string("input_761_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_7071_split_num_splits_0 = const()[name = string("op_7071_split_num_splits_0"), val = int32(2)];
            int32 var_7071_split_axis_0 = const()[name = string("op_7071_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_761_cast_fp16 = transpose(perm = input_761_perm_0, x = var_7068_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 1024, 188]> var_7071_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_7071_split_cast_fp16_1 = split(axis = var_7071_split_axis_0, num_splits = var_7071_split_num_splits_0, x = input_761_cast_fp16)[name = string("op_7071_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7071_split_1_sigmoid_cast_fp16 = sigmoid(x = var_7071_split_cast_fp16_1)[name = string("op_7071_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7071_cast_fp16 = mul(x = var_7071_split_cast_fp16_0, y = var_7071_split_1_sigmoid_cast_fp16)[name = string("op_7071_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_763_cast_fp16 = mul(x = var_7071_cast_fp16, y = var_2228_cast_fp16)[name = string("input_763_cast_fp16")];
            string input_765_pad_type_0 = const()[name = string("input_765_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_765_pad_0 = const()[name = string("input_765_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_765_groups_0 = const()[name = string("input_765_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_765_strides_0 = const()[name = string("input_765_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_765_dilations_0 = const()[name = string("input_765_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164399168)))];
            tensor<fp16, [1024]> const_78_to_fp16 = const()[name = string("const_78_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164417664)))];
            tensor<fp16, [1, 1024, 188]> input_767_cast_fp16 = conv(bias = const_78_to_fp16, dilations = input_765_dilations_0, groups = input_765_groups_0, pad = input_765_pad_0, pad_type = input_765_pad_type_0, strides = input_765_strides_0, weight = const_77_to_fp16, x = input_763_cast_fp16)[name = string("input_767_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_113_cast_fp16 = silu(x = input_767_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<int32, [3]> var_7111_perm_0 = const()[name = string("op_7111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_18_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164419776))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164681984))))[name = string("e_layers_18_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_7111_cast_fp16 = transpose(perm = var_7111_perm_0, x = x_113_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 188, 1024]> var_7113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_conv_pointwise_conv2_weight_to_fp16_lut, x = var_7111_cast_fp16)[name = string("op_7113_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_769_cast_fp16 = add(x = input_759_cast_fp16, y = var_7113_cast_fp16)[name = string("input_769_cast_fp16")];
            tensor<int32, [1]> input_771_axes_0 = const()[name = string("input_771_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164747584)))];
            tensor<fp16, [1024]> e_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164749696)))];
            fp16 var_7121_to_fp16 = const()[name = string("op_7121_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_771_cast_fp16 = layer_norm(axes = input_771_axes_0, beta = e_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_7121_to_fp16, gamma = e_layers_18_norm_feed_forward2_weight_to_fp16, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_18_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164751808))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165800448))))[name = string("e_layers_18_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_18_feed_forward2_linear1_weight_to_fp16_lut, x = input_771_cast_fp16)[name = string("linear_170_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_775_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_775_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_18_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166062656))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167111296))))[name = string("e_layers_18_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_18_feed_forward2_linear2_weight_to_fp16_lut, x = input_775_cast_fp16)[name = string("linear_171_cast_fp16")];
            fp16 var_7138_to_fp16 = const()[name = string("op_7138_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7139_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_7138_to_fp16)[name = string("op_7139_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_779_cast_fp16 = add(x = input_769_cast_fp16, y = var_7139_cast_fp16)[name = string("input_779_cast_fp16")];
            tensor<int32, [1]> input_781_axes_0 = const()[name = string("input_781_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_18_norm_out_weight_to_fp16 = const()[name = string("e_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167373504)))];
            tensor<fp16, [1024]> e_layers_18_norm_out_bias_to_fp16 = const()[name = string("e_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167375616)))];
            fp16 var_7143_to_fp16 = const()[name = string("op_7143_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_781_cast_fp16 = layer_norm(axes = input_781_axes_0, beta = e_layers_18_norm_out_bias_to_fp16, epsilon = var_7143_to_fp16, gamma = e_layers_18_norm_out_weight_to_fp16, x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
            tensor<int32, [1]> input_783_axes_0 = const()[name = string("input_783_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167377728)))];
            tensor<fp16, [1024]> e_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167379840)))];
            fp16 var_7150_to_fp16 = const()[name = string("op_7150_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_783_cast_fp16 = layer_norm(axes = input_783_axes_0, beta = e_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_7150_to_fp16, gamma = e_layers_19_norm_feed_forward1_weight_to_fp16, x = input_781_cast_fp16)[name = string("input_783_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_19_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167381952))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168430592))))[name = string("e_layers_19_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_19_feed_forward1_linear1_weight_to_fp16_lut, x = input_783_cast_fp16)[name = string("linear_172_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_787_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_787_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_19_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168692800))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169741440))))[name = string("e_layers_19_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_feed_forward1_linear2_weight_to_fp16_lut, x = input_787_cast_fp16)[name = string("linear_173_cast_fp16")];
            fp16 var_7167_to_fp16 = const()[name = string("op_7167_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7168_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_7167_to_fp16)[name = string("op_7168_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_791_cast_fp16 = add(x = input_781_cast_fp16, y = var_7168_cast_fp16)[name = string("input_791_cast_fp16")];
            tensor<int32, [1]> input_793_axes_0 = const()[name = string("input_793_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170003648)))];
            tensor<fp16, [1024]> e_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170005760)))];
            fp16 var_7172_to_fp16 = const()[name = string("op_7172_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_793_cast_fp16 = layer_norm(axes = input_793_axes_0, beta = e_layers_19_norm_self_att_bias_to_fp16, epsilon = var_7172_to_fp16, gamma = e_layers_19_norm_self_att_weight_to_fp16, x = input_791_cast_fp16)[name = string("input_793_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_19_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170007872))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170270080))))[name = string("e_layers_19_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_self_attn_q_proj_weight_to_fp16_lut, x = input_793_cast_fp16)[name = string("linear_174_cast_fp16")];
            tensor<int32, [4]> var_7185 = const()[name = string("op_7185"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7186_cast_fp16 = reshape(shape = var_7185, x = linear_174_cast_fp16)[name = string("op_7186_cast_fp16")];
            tensor<int32, [4]> q_39_perm_0 = const()[name = string("q_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_19_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170335680))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170597888))))[name = string("e_layers_19_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_self_attn_k_proj_weight_to_fp16_lut, x = input_793_cast_fp16)[name = string("linear_175_cast_fp16")];
            tensor<int32, [4]> var_7197 = const()[name = string("op_7197"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7198_cast_fp16 = reshape(shape = var_7197, x = linear_175_cast_fp16)[name = string("op_7198_cast_fp16")];
            tensor<int32, [4]> transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_19_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170663488))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170925696))))[name = string("e_layers_19_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_self_attn_v_proj_weight_to_fp16_lut, x = input_793_cast_fp16)[name = string("linear_176_cast_fp16")];
            tensor<int32, [4]> var_7209 = const()[name = string("op_7209"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7210_cast_fp16 = reshape(shape = var_7209, x = linear_176_cast_fp16)[name = string("op_7210_cast_fp16")];
            tensor<int32, [4]> v_39_perm_0 = const()[name = string("v_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_7234_to_fp16 = const()[name = string("op_7234_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170991296)))];
            tensor<fp16, [1, 8, 188, 128]> q_39_cast_fp16 = transpose(perm = q_39_perm_0, x = var_7186_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 8, 188, 128]> var_7236_cast_fp16 = add(x = q_39_cast_fp16, y = var_7234_to_fp16)[name = string("op_7236_cast_fp16")];
            bool input_795_transpose_x_0 = const()[name = string("input_795_transpose_x_0"), val = bool(false)];
            bool input_795_transpose_y_0 = const()[name = string("input_795_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_39_to_fp16 = const()[name = string("p_39_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170993408)))];
            tensor<fp16, [1, 8, 188, 375]> input_795_cast_fp16 = matmul(transpose_x = input_795_transpose_x_0, transpose_y = input_795_transpose_y_0, x = var_7236_cast_fp16, y = p_39_to_fp16)[name = string("input_795_cast_fp16")];
            tensor<int32, [8]> var_7243_pad_0 = const()[name = string("op_7243_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_7243_mode_0 = const()[name = string("op_7243_mode_0"), val = string("constant")];
            fp16 const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_7243_cast_fp16 = pad(constant_val = const_36_to_fp16, mode = var_7243_mode_0, pad = var_7243_pad_0, x = input_795_cast_fp16)[name = string("op_7243_cast_fp16")];
            tensor<int32, [4]> var_7248 = const()[name = string("op_7248"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_7249_cast_fp16 = reshape(shape = var_7248, x = var_7243_cast_fp16)[name = string("op_7249_cast_fp16")];
            tensor<int32, [4]> var_7264_begin_0 = const()[name = string("op_7264_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_7264_end_0 = const()[name = string("op_7264_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_7264_end_mask_0 = const()[name = string("op_7264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_7264_cast_fp16 = slice_by_index(begin = var_7264_begin_0, end = var_7264_end_0, end_mask = var_7264_end_mask_0, x = var_7249_cast_fp16)[name = string("op_7264_cast_fp16")];
            tensor<int32, [4]> var_7269 = const()[name = string("op_7269"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_7270_cast_fp16 = reshape(shape = var_7269, x = var_7264_cast_fp16)[name = string("op_7270_cast_fp16")];
            tensor<int32, [4]> bd_39_begin_0 = const()[name = string("bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_39_end_0 = const()[name = string("bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_39_end_mask_0 = const()[name = string("bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_39_cast_fp16 = slice_by_index(begin = bd_39_begin_0, end = bd_39_end_0, end_mask = bd_39_end_mask_0, x = var_7270_cast_fp16)[name = string("bd_39_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_7281_to_fp16 = const()[name = string("op_7281_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171761472)))];
            tensor<fp16, [1, 8, 188, 128]> var_7283_cast_fp16 = add(x = q_39_cast_fp16, y = var_7281_to_fp16)[name = string("op_7283_cast_fp16")];
            bool ac_39_transpose_x_0 = const()[name = string("ac_39_transpose_x_0"), val = bool(false)];
            bool ac_39_transpose_y_0 = const()[name = string("ac_39_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_67_cast_fp16 = transpose(perm = transpose_67_perm_0, x = var_7198_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 8, 188, 188]> ac_39_cast_fp16 = matmul(transpose_x = ac_39_transpose_x_0, transpose_y = ac_39_transpose_y_0, x = var_7283_cast_fp16, y = transpose_67_cast_fp16)[name = string("ac_39_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7289_cast_fp16 = add(x = ac_39_cast_fp16, y = bd_39_cast_fp16)[name = string("op_7289_cast_fp16")];
            fp16 var_7290_to_fp16 = const()[name = string("op_7290_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_7291_cast_fp16 = mul(x = var_7289_cast_fp16, y = var_7290_to_fp16)[name = string("op_7291_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7293_cast_fp16 = add(x = var_7291_cast_fp16, y = bias_15_cast_fp16)[name = string("op_7293_cast_fp16")];
            int32 var_7294 = const()[name = string("op_7294"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_39_cast_fp16 = softmax(axis = var_7294, x = var_7293_cast_fp16)[name = string("w_39_cast_fp16")];
            bool var_7297_transpose_x_0 = const()[name = string("op_7297_transpose_x_0"), val = bool(false)];
            bool var_7297_transpose_y_0 = const()[name = string("op_7297_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_39_cast_fp16 = transpose(perm = v_39_perm_0, x = var_7210_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 8, 188, 128]> var_7297_cast_fp16 = matmul(transpose_x = var_7297_transpose_x_0, transpose_y = var_7297_transpose_y_0, x = w_39_cast_fp16, y = v_39_cast_fp16)[name = string("op_7297_cast_fp16")];
            tensor<int32, [4]> var_7300_perm_0 = const()[name = string("op_7300_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7304 = const()[name = string("op_7304"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_7300_cast_fp16 = transpose(perm = var_7300_perm_0, x = var_7297_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 188, 1024]> input_797_cast_fp16 = reshape(shape = var_7304, x = var_7300_cast_fp16)[name = string("input_797_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_19_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171763584))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172025792))))[name = string("e_layers_19_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_self_attn_o_proj_weight_to_fp16_lut, x = input_797_cast_fp16)[name = string("linear_178_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_799_cast_fp16 = add(x = input_791_cast_fp16, y = linear_178_cast_fp16)[name = string("input_799_cast_fp16")];
            tensor<int32, [1]> x_115_axes_0 = const()[name = string("x_115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_19_norm_conv_weight_to_fp16 = const()[name = string("e_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172091392)))];
            tensor<fp16, [1024]> e_layers_19_norm_conv_bias_to_fp16 = const()[name = string("e_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172093504)))];
            fp16 var_7312_to_fp16 = const()[name = string("op_7312_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_115_cast_fp16 = layer_norm(axes = x_115_axes_0, beta = e_layers_19_norm_conv_bias_to_fp16, epsilon = var_7312_to_fp16, gamma = e_layers_19_norm_conv_weight_to_fp16, x = input_799_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_19_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172095616))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172619968))))[name = string("e_layers_19_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_7326_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_19_conv_pointwise_conv1_weight_to_fp16_lut, x = x_115_cast_fp16)[name = string("op_7326_cast_fp16")];
            tensor<int32, [3]> input_801_perm_0 = const()[name = string("input_801_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_7329_split_num_splits_0 = const()[name = string("op_7329_split_num_splits_0"), val = int32(2)];
            int32 var_7329_split_axis_0 = const()[name = string("op_7329_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_801_cast_fp16 = transpose(perm = input_801_perm_0, x = var_7326_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 1024, 188]> var_7329_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_7329_split_cast_fp16_1 = split(axis = var_7329_split_axis_0, num_splits = var_7329_split_num_splits_0, x = input_801_cast_fp16)[name = string("op_7329_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7329_split_1_sigmoid_cast_fp16 = sigmoid(x = var_7329_split_cast_fp16_1)[name = string("op_7329_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7329_cast_fp16 = mul(x = var_7329_split_cast_fp16_0, y = var_7329_split_1_sigmoid_cast_fp16)[name = string("op_7329_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_803_cast_fp16 = mul(x = var_7329_cast_fp16, y = var_2228_cast_fp16)[name = string("input_803_cast_fp16")];
            string input_805_pad_type_0 = const()[name = string("input_805_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_805_pad_0 = const()[name = string("input_805_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_805_groups_0 = const()[name = string("input_805_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_805_strides_0 = const()[name = string("input_805_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_805_dilations_0 = const()[name = string("input_805_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_79_to_fp16 = const()[name = string("const_79_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172751104)))];
            tensor<fp16, [1024]> const_80_to_fp16 = const()[name = string("const_80_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172769600)))];
            tensor<fp16, [1, 1024, 188]> input_807_cast_fp16 = conv(bias = const_80_to_fp16, dilations = input_805_dilations_0, groups = input_805_groups_0, pad = input_805_pad_0, pad_type = input_805_pad_type_0, strides = input_805_strides_0, weight = const_79_to_fp16, x = input_803_cast_fp16)[name = string("input_807_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_119_cast_fp16 = silu(x = input_807_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<int32, [3]> var_7369_perm_0 = const()[name = string("op_7369_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_19_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172771712))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173033920))))[name = string("e_layers_19_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_7369_cast_fp16 = transpose(perm = var_7369_perm_0, x = x_119_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 188, 1024]> var_7371_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_conv_pointwise_conv2_weight_to_fp16_lut, x = var_7369_cast_fp16)[name = string("op_7371_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_809_cast_fp16 = add(x = input_799_cast_fp16, y = var_7371_cast_fp16)[name = string("input_809_cast_fp16")];
            tensor<int32, [1]> input_811_axes_0 = const()[name = string("input_811_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173099520)))];
            tensor<fp16, [1024]> e_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173101632)))];
            fp16 var_7379_to_fp16 = const()[name = string("op_7379_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_811_cast_fp16 = layer_norm(axes = input_811_axes_0, beta = e_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_7379_to_fp16, gamma = e_layers_19_norm_feed_forward2_weight_to_fp16, x = input_809_cast_fp16)[name = string("input_811_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_19_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173103744))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174152384))))[name = string("e_layers_19_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_19_feed_forward2_linear1_weight_to_fp16_lut, x = input_811_cast_fp16)[name = string("linear_179_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_815_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_815_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_19_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174414592))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175463232))))[name = string("e_layers_19_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_19_feed_forward2_linear2_weight_to_fp16_lut, x = input_815_cast_fp16)[name = string("linear_180_cast_fp16")];
            fp16 var_7396_to_fp16 = const()[name = string("op_7396_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7397_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_7396_to_fp16)[name = string("op_7397_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_819_cast_fp16 = add(x = input_809_cast_fp16, y = var_7397_cast_fp16)[name = string("input_819_cast_fp16")];
            tensor<int32, [1]> input_821_axes_0 = const()[name = string("input_821_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_19_norm_out_weight_to_fp16 = const()[name = string("e_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175725440)))];
            tensor<fp16, [1024]> e_layers_19_norm_out_bias_to_fp16 = const()[name = string("e_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175727552)))];
            fp16 var_7401_to_fp16 = const()[name = string("op_7401_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_821_cast_fp16 = layer_norm(axes = input_821_axes_0, beta = e_layers_19_norm_out_bias_to_fp16, epsilon = var_7401_to_fp16, gamma = e_layers_19_norm_out_weight_to_fp16, x = input_819_cast_fp16)[name = string("input_821_cast_fp16")];
            tensor<int32, [1]> input_823_axes_0 = const()[name = string("input_823_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175729664)))];
            tensor<fp16, [1024]> e_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175731776)))];
            fp16 var_7408_to_fp16 = const()[name = string("op_7408_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_823_cast_fp16 = layer_norm(axes = input_823_axes_0, beta = e_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_7408_to_fp16, gamma = e_layers_20_norm_feed_forward1_weight_to_fp16, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_20_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175733888))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176782528))))[name = string("e_layers_20_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_20_feed_forward1_linear1_weight_to_fp16_lut, x = input_823_cast_fp16)[name = string("linear_181_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_827_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_827_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_20_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177044736))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178093376))))[name = string("e_layers_20_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_feed_forward1_linear2_weight_to_fp16_lut, x = input_827_cast_fp16)[name = string("linear_182_cast_fp16")];
            fp16 var_7425_to_fp16 = const()[name = string("op_7425_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7426_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_7425_to_fp16)[name = string("op_7426_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_831_cast_fp16 = add(x = input_821_cast_fp16, y = var_7426_cast_fp16)[name = string("input_831_cast_fp16")];
            tensor<int32, [1]> input_833_axes_0 = const()[name = string("input_833_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178355584)))];
            tensor<fp16, [1024]> e_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178357696)))];
            fp16 var_7430_to_fp16 = const()[name = string("op_7430_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_833_cast_fp16 = layer_norm(axes = input_833_axes_0, beta = e_layers_20_norm_self_att_bias_to_fp16, epsilon = var_7430_to_fp16, gamma = e_layers_20_norm_self_att_weight_to_fp16, x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_20_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178359808))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178622016))))[name = string("e_layers_20_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_self_attn_q_proj_weight_to_fp16_lut, x = input_833_cast_fp16)[name = string("linear_183_cast_fp16")];
            tensor<int32, [4]> var_7443 = const()[name = string("op_7443"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7444_cast_fp16 = reshape(shape = var_7443, x = linear_183_cast_fp16)[name = string("op_7444_cast_fp16")];
            tensor<int32, [4]> q_41_perm_0 = const()[name = string("q_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_20_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178687616))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178949824))))[name = string("e_layers_20_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_self_attn_k_proj_weight_to_fp16_lut, x = input_833_cast_fp16)[name = string("linear_184_cast_fp16")];
            tensor<int32, [4]> var_7455 = const()[name = string("op_7455"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7456_cast_fp16 = reshape(shape = var_7455, x = linear_184_cast_fp16)[name = string("op_7456_cast_fp16")];
            tensor<int32, [4]> transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_20_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179015424))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179277632))))[name = string("e_layers_20_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_self_attn_v_proj_weight_to_fp16_lut, x = input_833_cast_fp16)[name = string("linear_185_cast_fp16")];
            tensor<int32, [4]> var_7467 = const()[name = string("op_7467"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7468_cast_fp16 = reshape(shape = var_7467, x = linear_185_cast_fp16)[name = string("op_7468_cast_fp16")];
            tensor<int32, [4]> v_41_perm_0 = const()[name = string("v_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_7492_to_fp16 = const()[name = string("op_7492_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179343232)))];
            tensor<fp16, [1, 8, 188, 128]> q_41_cast_fp16 = transpose(perm = q_41_perm_0, x = var_7444_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 8, 188, 128]> var_7494_cast_fp16 = add(x = q_41_cast_fp16, y = var_7492_to_fp16)[name = string("op_7494_cast_fp16")];
            bool input_835_transpose_x_0 = const()[name = string("input_835_transpose_x_0"), val = bool(false)];
            bool input_835_transpose_y_0 = const()[name = string("input_835_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_41_to_fp16 = const()[name = string("p_41_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179345344)))];
            tensor<fp16, [1, 8, 188, 375]> input_835_cast_fp16 = matmul(transpose_x = input_835_transpose_x_0, transpose_y = input_835_transpose_y_0, x = var_7494_cast_fp16, y = p_41_to_fp16)[name = string("input_835_cast_fp16")];
            tensor<int32, [8]> var_7501_pad_0 = const()[name = string("op_7501_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_7501_mode_0 = const()[name = string("op_7501_mode_0"), val = string("constant")];
            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_7501_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = var_7501_mode_0, pad = var_7501_pad_0, x = input_835_cast_fp16)[name = string("op_7501_cast_fp16")];
            tensor<int32, [4]> var_7506 = const()[name = string("op_7506"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_7507_cast_fp16 = reshape(shape = var_7506, x = var_7501_cast_fp16)[name = string("op_7507_cast_fp16")];
            tensor<int32, [4]> var_7522_begin_0 = const()[name = string("op_7522_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_7522_end_0 = const()[name = string("op_7522_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_7522_end_mask_0 = const()[name = string("op_7522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_7522_cast_fp16 = slice_by_index(begin = var_7522_begin_0, end = var_7522_end_0, end_mask = var_7522_end_mask_0, x = var_7507_cast_fp16)[name = string("op_7522_cast_fp16")];
            tensor<int32, [4]> var_7527 = const()[name = string("op_7527"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_7528_cast_fp16 = reshape(shape = var_7527, x = var_7522_cast_fp16)[name = string("op_7528_cast_fp16")];
            tensor<int32, [4]> bd_41_begin_0 = const()[name = string("bd_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_41_end_0 = const()[name = string("bd_41_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_41_end_mask_0 = const()[name = string("bd_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_41_cast_fp16 = slice_by_index(begin = bd_41_begin_0, end = bd_41_end_0, end_mask = bd_41_end_mask_0, x = var_7528_cast_fp16)[name = string("bd_41_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_7539_to_fp16 = const()[name = string("op_7539_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180113408)))];
            tensor<fp16, [1, 8, 188, 128]> var_7541_cast_fp16 = add(x = q_41_cast_fp16, y = var_7539_to_fp16)[name = string("op_7541_cast_fp16")];
            bool ac_41_transpose_x_0 = const()[name = string("ac_41_transpose_x_0"), val = bool(false)];
            bool ac_41_transpose_y_0 = const()[name = string("ac_41_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_68_cast_fp16 = transpose(perm = transpose_68_perm_0, x = var_7456_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 8, 188, 188]> ac_41_cast_fp16 = matmul(transpose_x = ac_41_transpose_x_0, transpose_y = ac_41_transpose_y_0, x = var_7541_cast_fp16, y = transpose_68_cast_fp16)[name = string("ac_41_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7547_cast_fp16 = add(x = ac_41_cast_fp16, y = bd_41_cast_fp16)[name = string("op_7547_cast_fp16")];
            fp16 var_7548_to_fp16 = const()[name = string("op_7548_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_7549_cast_fp16 = mul(x = var_7547_cast_fp16, y = var_7548_to_fp16)[name = string("op_7549_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7551_cast_fp16 = add(x = var_7549_cast_fp16, y = bias_15_cast_fp16)[name = string("op_7551_cast_fp16")];
            int32 var_7552 = const()[name = string("op_7552"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_41_cast_fp16 = softmax(axis = var_7552, x = var_7551_cast_fp16)[name = string("w_41_cast_fp16")];
            bool var_7555_transpose_x_0 = const()[name = string("op_7555_transpose_x_0"), val = bool(false)];
            bool var_7555_transpose_y_0 = const()[name = string("op_7555_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_41_cast_fp16 = transpose(perm = v_41_perm_0, x = var_7468_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 8, 188, 128]> var_7555_cast_fp16 = matmul(transpose_x = var_7555_transpose_x_0, transpose_y = var_7555_transpose_y_0, x = w_41_cast_fp16, y = v_41_cast_fp16)[name = string("op_7555_cast_fp16")];
            tensor<int32, [4]> var_7558_perm_0 = const()[name = string("op_7558_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7562 = const()[name = string("op_7562"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_7558_cast_fp16 = transpose(perm = var_7558_perm_0, x = var_7555_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 188, 1024]> input_837_cast_fp16 = reshape(shape = var_7562, x = var_7558_cast_fp16)[name = string("input_837_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_20_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180115520))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180377728))))[name = string("e_layers_20_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_self_attn_o_proj_weight_to_fp16_lut, x = input_837_cast_fp16)[name = string("linear_187_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_839_cast_fp16 = add(x = input_831_cast_fp16, y = linear_187_cast_fp16)[name = string("input_839_cast_fp16")];
            tensor<int32, [1]> x_121_axes_0 = const()[name = string("x_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_20_norm_conv_weight_to_fp16 = const()[name = string("e_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180443328)))];
            tensor<fp16, [1024]> e_layers_20_norm_conv_bias_to_fp16 = const()[name = string("e_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180445440)))];
            fp16 var_7570_to_fp16 = const()[name = string("op_7570_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_121_cast_fp16 = layer_norm(axes = x_121_axes_0, beta = e_layers_20_norm_conv_bias_to_fp16, epsilon = var_7570_to_fp16, gamma = e_layers_20_norm_conv_weight_to_fp16, x = input_839_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_20_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180447552))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180971904))))[name = string("e_layers_20_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_7584_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_20_conv_pointwise_conv1_weight_to_fp16_lut, x = x_121_cast_fp16)[name = string("op_7584_cast_fp16")];
            tensor<int32, [3]> input_841_perm_0 = const()[name = string("input_841_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_7587_split_num_splits_0 = const()[name = string("op_7587_split_num_splits_0"), val = int32(2)];
            int32 var_7587_split_axis_0 = const()[name = string("op_7587_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_841_cast_fp16 = transpose(perm = input_841_perm_0, x = var_7584_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 1024, 188]> var_7587_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_7587_split_cast_fp16_1 = split(axis = var_7587_split_axis_0, num_splits = var_7587_split_num_splits_0, x = input_841_cast_fp16)[name = string("op_7587_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7587_split_1_sigmoid_cast_fp16 = sigmoid(x = var_7587_split_cast_fp16_1)[name = string("op_7587_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7587_cast_fp16 = mul(x = var_7587_split_cast_fp16_0, y = var_7587_split_1_sigmoid_cast_fp16)[name = string("op_7587_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_843_cast_fp16 = mul(x = var_7587_cast_fp16, y = var_2228_cast_fp16)[name = string("input_843_cast_fp16")];
            string input_845_pad_type_0 = const()[name = string("input_845_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_845_pad_0 = const()[name = string("input_845_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_845_groups_0 = const()[name = string("input_845_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_845_strides_0 = const()[name = string("input_845_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_845_dilations_0 = const()[name = string("input_845_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_81_to_fp16 = const()[name = string("const_81_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181103040)))];
            tensor<fp16, [1024]> const_82_to_fp16 = const()[name = string("const_82_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181121536)))];
            tensor<fp16, [1, 1024, 188]> input_847_cast_fp16 = conv(bias = const_82_to_fp16, dilations = input_845_dilations_0, groups = input_845_groups_0, pad = input_845_pad_0, pad_type = input_845_pad_type_0, strides = input_845_strides_0, weight = const_81_to_fp16, x = input_843_cast_fp16)[name = string("input_847_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_125_cast_fp16 = silu(x = input_847_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<int32, [3]> var_7627_perm_0 = const()[name = string("op_7627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_20_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181123648))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181385856))))[name = string("e_layers_20_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_7627_cast_fp16 = transpose(perm = var_7627_perm_0, x = x_125_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 188, 1024]> var_7629_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_conv_pointwise_conv2_weight_to_fp16_lut, x = var_7627_cast_fp16)[name = string("op_7629_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_849_cast_fp16 = add(x = input_839_cast_fp16, y = var_7629_cast_fp16)[name = string("input_849_cast_fp16")];
            tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181451456)))];
            tensor<fp16, [1024]> e_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181453568)))];
            fp16 var_7637_to_fp16 = const()[name = string("op_7637_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = e_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_7637_to_fp16, gamma = e_layers_20_norm_feed_forward2_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_20_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181455680))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182504320))))[name = string("e_layers_20_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_20_feed_forward2_linear1_weight_to_fp16_lut, x = input_851_cast_fp16)[name = string("linear_188_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_855_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_855_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_20_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182766528))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183815168))))[name = string("e_layers_20_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_20_feed_forward2_linear2_weight_to_fp16_lut, x = input_855_cast_fp16)[name = string("linear_189_cast_fp16")];
            fp16 var_7654_to_fp16 = const()[name = string("op_7654_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7655_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_7654_to_fp16)[name = string("op_7655_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_859_cast_fp16 = add(x = input_849_cast_fp16, y = var_7655_cast_fp16)[name = string("input_859_cast_fp16")];
            tensor<int32, [1]> input_861_axes_0 = const()[name = string("input_861_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_20_norm_out_weight_to_fp16 = const()[name = string("e_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184077376)))];
            tensor<fp16, [1024]> e_layers_20_norm_out_bias_to_fp16 = const()[name = string("e_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184079488)))];
            fp16 var_7659_to_fp16 = const()[name = string("op_7659_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_861_cast_fp16 = layer_norm(axes = input_861_axes_0, beta = e_layers_20_norm_out_bias_to_fp16, epsilon = var_7659_to_fp16, gamma = e_layers_20_norm_out_weight_to_fp16, x = input_859_cast_fp16)[name = string("input_861_cast_fp16")];
            tensor<int32, [1]> input_863_axes_0 = const()[name = string("input_863_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184081600)))];
            tensor<fp16, [1024]> e_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184083712)))];
            fp16 var_7666_to_fp16 = const()[name = string("op_7666_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_863_cast_fp16 = layer_norm(axes = input_863_axes_0, beta = e_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_7666_to_fp16, gamma = e_layers_21_norm_feed_forward1_weight_to_fp16, x = input_861_cast_fp16)[name = string("input_863_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_21_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184085824))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185134464))))[name = string("e_layers_21_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_21_feed_forward1_linear1_weight_to_fp16_lut, x = input_863_cast_fp16)[name = string("linear_190_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_867_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_867_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_21_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185396672))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186445312))))[name = string("e_layers_21_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_feed_forward1_linear2_weight_to_fp16_lut, x = input_867_cast_fp16)[name = string("linear_191_cast_fp16")];
            fp16 var_7683_to_fp16 = const()[name = string("op_7683_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7684_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_7683_to_fp16)[name = string("op_7684_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_871_cast_fp16 = add(x = input_861_cast_fp16, y = var_7684_cast_fp16)[name = string("input_871_cast_fp16")];
            tensor<int32, [1]> input_873_axes_0 = const()[name = string("input_873_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186707520)))];
            tensor<fp16, [1024]> e_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186709632)))];
            fp16 var_7688_to_fp16 = const()[name = string("op_7688_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_873_cast_fp16 = layer_norm(axes = input_873_axes_0, beta = e_layers_21_norm_self_att_bias_to_fp16, epsilon = var_7688_to_fp16, gamma = e_layers_21_norm_self_att_weight_to_fp16, x = input_871_cast_fp16)[name = string("input_873_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_21_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186711744))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186973952))))[name = string("e_layers_21_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_self_attn_q_proj_weight_to_fp16_lut, x = input_873_cast_fp16)[name = string("linear_192_cast_fp16")];
            tensor<int32, [4]> var_7701 = const()[name = string("op_7701"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7702_cast_fp16 = reshape(shape = var_7701, x = linear_192_cast_fp16)[name = string("op_7702_cast_fp16")];
            tensor<int32, [4]> q_43_perm_0 = const()[name = string("q_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_21_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187039552))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187301760))))[name = string("e_layers_21_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_self_attn_k_proj_weight_to_fp16_lut, x = input_873_cast_fp16)[name = string("linear_193_cast_fp16")];
            tensor<int32, [4]> var_7713 = const()[name = string("op_7713"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7714_cast_fp16 = reshape(shape = var_7713, x = linear_193_cast_fp16)[name = string("op_7714_cast_fp16")];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_21_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187367360))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187629568))))[name = string("e_layers_21_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_self_attn_v_proj_weight_to_fp16_lut, x = input_873_cast_fp16)[name = string("linear_194_cast_fp16")];
            tensor<int32, [4]> var_7725 = const()[name = string("op_7725"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7726_cast_fp16 = reshape(shape = var_7725, x = linear_194_cast_fp16)[name = string("op_7726_cast_fp16")];
            tensor<int32, [4]> v_43_perm_0 = const()[name = string("v_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_7750_to_fp16 = const()[name = string("op_7750_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187695168)))];
            tensor<fp16, [1, 8, 188, 128]> q_43_cast_fp16 = transpose(perm = q_43_perm_0, x = var_7702_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 8, 188, 128]> var_7752_cast_fp16 = add(x = q_43_cast_fp16, y = var_7750_to_fp16)[name = string("op_7752_cast_fp16")];
            bool input_875_transpose_x_0 = const()[name = string("input_875_transpose_x_0"), val = bool(false)];
            bool input_875_transpose_y_0 = const()[name = string("input_875_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_43_to_fp16 = const()[name = string("p_43_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187697280)))];
            tensor<fp16, [1, 8, 188, 375]> input_875_cast_fp16 = matmul(transpose_x = input_875_transpose_x_0, transpose_y = input_875_transpose_y_0, x = var_7752_cast_fp16, y = p_43_to_fp16)[name = string("input_875_cast_fp16")];
            tensor<int32, [8]> var_7759_pad_0 = const()[name = string("op_7759_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_7759_mode_0 = const()[name = string("op_7759_mode_0"), val = string("constant")];
            fp16 const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_7759_cast_fp16 = pad(constant_val = const_38_to_fp16, mode = var_7759_mode_0, pad = var_7759_pad_0, x = input_875_cast_fp16)[name = string("op_7759_cast_fp16")];
            tensor<int32, [4]> var_7764 = const()[name = string("op_7764"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_7765_cast_fp16 = reshape(shape = var_7764, x = var_7759_cast_fp16)[name = string("op_7765_cast_fp16")];
            tensor<int32, [4]> var_7780_begin_0 = const()[name = string("op_7780_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_7780_end_0 = const()[name = string("op_7780_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_7780_end_mask_0 = const()[name = string("op_7780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_7780_cast_fp16 = slice_by_index(begin = var_7780_begin_0, end = var_7780_end_0, end_mask = var_7780_end_mask_0, x = var_7765_cast_fp16)[name = string("op_7780_cast_fp16")];
            tensor<int32, [4]> var_7785 = const()[name = string("op_7785"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_7786_cast_fp16 = reshape(shape = var_7785, x = var_7780_cast_fp16)[name = string("op_7786_cast_fp16")];
            tensor<int32, [4]> bd_43_begin_0 = const()[name = string("bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_43_end_0 = const()[name = string("bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_43_end_mask_0 = const()[name = string("bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_43_cast_fp16 = slice_by_index(begin = bd_43_begin_0, end = bd_43_end_0, end_mask = bd_43_end_mask_0, x = var_7786_cast_fp16)[name = string("bd_43_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_7797_to_fp16 = const()[name = string("op_7797_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188465344)))];
            tensor<fp16, [1, 8, 188, 128]> var_7799_cast_fp16 = add(x = q_43_cast_fp16, y = var_7797_to_fp16)[name = string("op_7799_cast_fp16")];
            bool ac_43_transpose_x_0 = const()[name = string("ac_43_transpose_x_0"), val = bool(false)];
            bool ac_43_transpose_y_0 = const()[name = string("ac_43_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_69_cast_fp16 = transpose(perm = transpose_69_perm_0, x = var_7714_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 8, 188, 188]> ac_43_cast_fp16 = matmul(transpose_x = ac_43_transpose_x_0, transpose_y = ac_43_transpose_y_0, x = var_7799_cast_fp16, y = transpose_69_cast_fp16)[name = string("ac_43_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7805_cast_fp16 = add(x = ac_43_cast_fp16, y = bd_43_cast_fp16)[name = string("op_7805_cast_fp16")];
            fp16 var_7806_to_fp16 = const()[name = string("op_7806_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_7807_cast_fp16 = mul(x = var_7805_cast_fp16, y = var_7806_to_fp16)[name = string("op_7807_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_7809_cast_fp16 = add(x = var_7807_cast_fp16, y = bias_15_cast_fp16)[name = string("op_7809_cast_fp16")];
            int32 var_7810 = const()[name = string("op_7810"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_43_cast_fp16 = softmax(axis = var_7810, x = var_7809_cast_fp16)[name = string("w_43_cast_fp16")];
            bool var_7813_transpose_x_0 = const()[name = string("op_7813_transpose_x_0"), val = bool(false)];
            bool var_7813_transpose_y_0 = const()[name = string("op_7813_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_43_cast_fp16 = transpose(perm = v_43_perm_0, x = var_7726_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 8, 188, 128]> var_7813_cast_fp16 = matmul(transpose_x = var_7813_transpose_x_0, transpose_y = var_7813_transpose_y_0, x = w_43_cast_fp16, y = v_43_cast_fp16)[name = string("op_7813_cast_fp16")];
            tensor<int32, [4]> var_7816_perm_0 = const()[name = string("op_7816_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7820 = const()[name = string("op_7820"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_7816_cast_fp16 = transpose(perm = var_7816_perm_0, x = var_7813_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 188, 1024]> input_877_cast_fp16 = reshape(shape = var_7820, x = var_7816_cast_fp16)[name = string("input_877_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_21_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188467456))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188729664))))[name = string("e_layers_21_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_self_attn_o_proj_weight_to_fp16_lut, x = input_877_cast_fp16)[name = string("linear_196_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_879_cast_fp16 = add(x = input_871_cast_fp16, y = linear_196_cast_fp16)[name = string("input_879_cast_fp16")];
            tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_21_norm_conv_weight_to_fp16 = const()[name = string("e_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188795264)))];
            tensor<fp16, [1024]> e_layers_21_norm_conv_bias_to_fp16 = const()[name = string("e_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188797376)))];
            fp16 var_7828_to_fp16 = const()[name = string("op_7828_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = e_layers_21_norm_conv_bias_to_fp16, epsilon = var_7828_to_fp16, gamma = e_layers_21_norm_conv_weight_to_fp16, x = input_879_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_21_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188799488))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189323840))))[name = string("e_layers_21_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_7842_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_21_conv_pointwise_conv1_weight_to_fp16_lut, x = x_127_cast_fp16)[name = string("op_7842_cast_fp16")];
            tensor<int32, [3]> input_881_perm_0 = const()[name = string("input_881_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_7845_split_num_splits_0 = const()[name = string("op_7845_split_num_splits_0"), val = int32(2)];
            int32 var_7845_split_axis_0 = const()[name = string("op_7845_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_881_cast_fp16 = transpose(perm = input_881_perm_0, x = var_7842_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 1024, 188]> var_7845_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_7845_split_cast_fp16_1 = split(axis = var_7845_split_axis_0, num_splits = var_7845_split_num_splits_0, x = input_881_cast_fp16)[name = string("op_7845_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7845_split_1_sigmoid_cast_fp16 = sigmoid(x = var_7845_split_cast_fp16_1)[name = string("op_7845_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_7845_cast_fp16 = mul(x = var_7845_split_cast_fp16_0, y = var_7845_split_1_sigmoid_cast_fp16)[name = string("op_7845_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_883_cast_fp16 = mul(x = var_7845_cast_fp16, y = var_2228_cast_fp16)[name = string("input_883_cast_fp16")];
            string input_885_pad_type_0 = const()[name = string("input_885_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_885_pad_0 = const()[name = string("input_885_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_885_groups_0 = const()[name = string("input_885_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_885_strides_0 = const()[name = string("input_885_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_885_dilations_0 = const()[name = string("input_885_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_83_to_fp16 = const()[name = string("const_83_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189454976)))];
            tensor<fp16, [1024]> const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189473472)))];
            tensor<fp16, [1, 1024, 188]> input_887_cast_fp16 = conv(bias = const_84_to_fp16, dilations = input_885_dilations_0, groups = input_885_groups_0, pad = input_885_pad_0, pad_type = input_885_pad_type_0, strides = input_885_strides_0, weight = const_83_to_fp16, x = input_883_cast_fp16)[name = string("input_887_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_131_cast_fp16 = silu(x = input_887_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<int32, [3]> var_7885_perm_0 = const()[name = string("op_7885_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_21_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189475584))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189737792))))[name = string("e_layers_21_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_7885_cast_fp16 = transpose(perm = var_7885_perm_0, x = x_131_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 188, 1024]> var_7887_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_conv_pointwise_conv2_weight_to_fp16_lut, x = var_7885_cast_fp16)[name = string("op_7887_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_889_cast_fp16 = add(x = input_879_cast_fp16, y = var_7887_cast_fp16)[name = string("input_889_cast_fp16")];
            tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189803392)))];
            tensor<fp16, [1024]> e_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189805504)))];
            fp16 var_7895_to_fp16 = const()[name = string("op_7895_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = e_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_7895_to_fp16, gamma = e_layers_21_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_21_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189807616))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190856256))))[name = string("e_layers_21_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_21_feed_forward2_linear1_weight_to_fp16_lut, x = input_891_cast_fp16)[name = string("linear_197_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_895_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_895_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_21_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191118464))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192167104))))[name = string("e_layers_21_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_21_feed_forward2_linear2_weight_to_fp16_lut, x = input_895_cast_fp16)[name = string("linear_198_cast_fp16")];
            fp16 var_7912_to_fp16 = const()[name = string("op_7912_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7913_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_7912_to_fp16)[name = string("op_7913_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_899_cast_fp16 = add(x = input_889_cast_fp16, y = var_7913_cast_fp16)[name = string("input_899_cast_fp16")];
            tensor<int32, [1]> input_901_axes_0 = const()[name = string("input_901_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_21_norm_out_weight_to_fp16 = const()[name = string("e_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192429312)))];
            tensor<fp16, [1024]> e_layers_21_norm_out_bias_to_fp16 = const()[name = string("e_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192431424)))];
            fp16 var_7917_to_fp16 = const()[name = string("op_7917_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_901_cast_fp16 = layer_norm(axes = input_901_axes_0, beta = e_layers_21_norm_out_bias_to_fp16, epsilon = var_7917_to_fp16, gamma = e_layers_21_norm_out_weight_to_fp16, x = input_899_cast_fp16)[name = string("input_901_cast_fp16")];
            tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192433536)))];
            tensor<fp16, [1024]> e_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192435648)))];
            fp16 var_7924_to_fp16 = const()[name = string("op_7924_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = e_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_7924_to_fp16, gamma = e_layers_22_norm_feed_forward1_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_22_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192437760))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193486400))))[name = string("e_layers_22_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_22_feed_forward1_linear1_weight_to_fp16_lut, x = input_903_cast_fp16)[name = string("linear_199_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_907_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_907_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_22_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193748608))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194797248))))[name = string("e_layers_22_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_feed_forward1_linear2_weight_to_fp16_lut, x = input_907_cast_fp16)[name = string("linear_200_cast_fp16")];
            fp16 var_7941_to_fp16 = const()[name = string("op_7941_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_7942_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_7941_to_fp16)[name = string("op_7942_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_911_cast_fp16 = add(x = input_901_cast_fp16, y = var_7942_cast_fp16)[name = string("input_911_cast_fp16")];
            tensor<int32, [1]> input_913_axes_0 = const()[name = string("input_913_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195059456)))];
            tensor<fp16, [1024]> e_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195061568)))];
            fp16 var_7946_to_fp16 = const()[name = string("op_7946_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_913_cast_fp16 = layer_norm(axes = input_913_axes_0, beta = e_layers_22_norm_self_att_bias_to_fp16, epsilon = var_7946_to_fp16, gamma = e_layers_22_norm_self_att_weight_to_fp16, x = input_911_cast_fp16)[name = string("input_913_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_22_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195063680))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195325888))))[name = string("e_layers_22_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_self_attn_q_proj_weight_to_fp16_lut, x = input_913_cast_fp16)[name = string("linear_201_cast_fp16")];
            tensor<int32, [4]> var_7959 = const()[name = string("op_7959"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7960_cast_fp16 = reshape(shape = var_7959, x = linear_201_cast_fp16)[name = string("op_7960_cast_fp16")];
            tensor<int32, [4]> q_45_perm_0 = const()[name = string("q_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_22_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195391488))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195653696))))[name = string("e_layers_22_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_self_attn_k_proj_weight_to_fp16_lut, x = input_913_cast_fp16)[name = string("linear_202_cast_fp16")];
            tensor<int32, [4]> var_7971 = const()[name = string("op_7971"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7972_cast_fp16 = reshape(shape = var_7971, x = linear_202_cast_fp16)[name = string("op_7972_cast_fp16")];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_22_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195719296))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195981504))))[name = string("e_layers_22_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_self_attn_v_proj_weight_to_fp16_lut, x = input_913_cast_fp16)[name = string("linear_203_cast_fp16")];
            tensor<int32, [4]> var_7983 = const()[name = string("op_7983"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_7984_cast_fp16 = reshape(shape = var_7983, x = linear_203_cast_fp16)[name = string("op_7984_cast_fp16")];
            tensor<int32, [4]> v_45_perm_0 = const()[name = string("v_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_8008_to_fp16 = const()[name = string("op_8008_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196047104)))];
            tensor<fp16, [1, 8, 188, 128]> q_45_cast_fp16 = transpose(perm = q_45_perm_0, x = var_7960_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 8, 188, 128]> var_8010_cast_fp16 = add(x = q_45_cast_fp16, y = var_8008_to_fp16)[name = string("op_8010_cast_fp16")];
            bool input_915_transpose_x_0 = const()[name = string("input_915_transpose_x_0"), val = bool(false)];
            bool input_915_transpose_y_0 = const()[name = string("input_915_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_45_to_fp16 = const()[name = string("p_45_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196049216)))];
            tensor<fp16, [1, 8, 188, 375]> input_915_cast_fp16 = matmul(transpose_x = input_915_transpose_x_0, transpose_y = input_915_transpose_y_0, x = var_8010_cast_fp16, y = p_45_to_fp16)[name = string("input_915_cast_fp16")];
            tensor<int32, [8]> var_8017_pad_0 = const()[name = string("op_8017_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_8017_mode_0 = const()[name = string("op_8017_mode_0"), val = string("constant")];
            fp16 const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_8017_cast_fp16 = pad(constant_val = const_39_to_fp16, mode = var_8017_mode_0, pad = var_8017_pad_0, x = input_915_cast_fp16)[name = string("op_8017_cast_fp16")];
            tensor<int32, [4]> var_8022 = const()[name = string("op_8022"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_8023_cast_fp16 = reshape(shape = var_8022, x = var_8017_cast_fp16)[name = string("op_8023_cast_fp16")];
            tensor<int32, [4]> var_8038_begin_0 = const()[name = string("op_8038_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_8038_end_0 = const()[name = string("op_8038_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_8038_end_mask_0 = const()[name = string("op_8038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_8038_cast_fp16 = slice_by_index(begin = var_8038_begin_0, end = var_8038_end_0, end_mask = var_8038_end_mask_0, x = var_8023_cast_fp16)[name = string("op_8038_cast_fp16")];
            tensor<int32, [4]> var_8043 = const()[name = string("op_8043"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_8044_cast_fp16 = reshape(shape = var_8043, x = var_8038_cast_fp16)[name = string("op_8044_cast_fp16")];
            tensor<int32, [4]> bd_45_begin_0 = const()[name = string("bd_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_45_end_0 = const()[name = string("bd_45_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_45_end_mask_0 = const()[name = string("bd_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_45_cast_fp16 = slice_by_index(begin = bd_45_begin_0, end = bd_45_end_0, end_mask = bd_45_end_mask_0, x = var_8044_cast_fp16)[name = string("bd_45_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_8055_to_fp16 = const()[name = string("op_8055_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196817280)))];
            tensor<fp16, [1, 8, 188, 128]> var_8057_cast_fp16 = add(x = q_45_cast_fp16, y = var_8055_to_fp16)[name = string("op_8057_cast_fp16")];
            bool ac_45_transpose_x_0 = const()[name = string("ac_45_transpose_x_0"), val = bool(false)];
            bool ac_45_transpose_y_0 = const()[name = string("ac_45_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_70_cast_fp16 = transpose(perm = transpose_70_perm_0, x = var_7972_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 8, 188, 188]> ac_45_cast_fp16 = matmul(transpose_x = ac_45_transpose_x_0, transpose_y = ac_45_transpose_y_0, x = var_8057_cast_fp16, y = transpose_70_cast_fp16)[name = string("ac_45_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_8063_cast_fp16 = add(x = ac_45_cast_fp16, y = bd_45_cast_fp16)[name = string("op_8063_cast_fp16")];
            fp16 var_8064_to_fp16 = const()[name = string("op_8064_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_8065_cast_fp16 = mul(x = var_8063_cast_fp16, y = var_8064_to_fp16)[name = string("op_8065_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_8067_cast_fp16 = add(x = var_8065_cast_fp16, y = bias_15_cast_fp16)[name = string("op_8067_cast_fp16")];
            int32 var_8068 = const()[name = string("op_8068"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_45_cast_fp16 = softmax(axis = var_8068, x = var_8067_cast_fp16)[name = string("w_45_cast_fp16")];
            bool var_8071_transpose_x_0 = const()[name = string("op_8071_transpose_x_0"), val = bool(false)];
            bool var_8071_transpose_y_0 = const()[name = string("op_8071_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_45_cast_fp16 = transpose(perm = v_45_perm_0, x = var_7984_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 8, 188, 128]> var_8071_cast_fp16 = matmul(transpose_x = var_8071_transpose_x_0, transpose_y = var_8071_transpose_y_0, x = w_45_cast_fp16, y = v_45_cast_fp16)[name = string("op_8071_cast_fp16")];
            tensor<int32, [4]> var_8074_perm_0 = const()[name = string("op_8074_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_8078 = const()[name = string("op_8078"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_8074_cast_fp16 = transpose(perm = var_8074_perm_0, x = var_8071_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 188, 1024]> input_917_cast_fp16 = reshape(shape = var_8078, x = var_8074_cast_fp16)[name = string("input_917_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_22_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196819392))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197081600))))[name = string("e_layers_22_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_self_attn_o_proj_weight_to_fp16_lut, x = input_917_cast_fp16)[name = string("linear_205_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_919_cast_fp16 = add(x = input_911_cast_fp16, y = linear_205_cast_fp16)[name = string("input_919_cast_fp16")];
            tensor<int32, [1]> x_133_axes_0 = const()[name = string("x_133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_22_norm_conv_weight_to_fp16 = const()[name = string("e_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197147200)))];
            tensor<fp16, [1024]> e_layers_22_norm_conv_bias_to_fp16 = const()[name = string("e_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197149312)))];
            fp16 var_8086_to_fp16 = const()[name = string("op_8086_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_133_cast_fp16 = layer_norm(axes = x_133_axes_0, beta = e_layers_22_norm_conv_bias_to_fp16, epsilon = var_8086_to_fp16, gamma = e_layers_22_norm_conv_weight_to_fp16, x = input_919_cast_fp16)[name = string("x_133_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_22_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197151424))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197675776))))[name = string("e_layers_22_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_8100_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_22_conv_pointwise_conv1_weight_to_fp16_lut, x = x_133_cast_fp16)[name = string("op_8100_cast_fp16")];
            tensor<int32, [3]> input_921_perm_0 = const()[name = string("input_921_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_8103_split_num_splits_0 = const()[name = string("op_8103_split_num_splits_0"), val = int32(2)];
            int32 var_8103_split_axis_0 = const()[name = string("op_8103_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_921_cast_fp16 = transpose(perm = input_921_perm_0, x = var_8100_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 1024, 188]> var_8103_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_8103_split_cast_fp16_1 = split(axis = var_8103_split_axis_0, num_splits = var_8103_split_num_splits_0, x = input_921_cast_fp16)[name = string("op_8103_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_8103_split_1_sigmoid_cast_fp16 = sigmoid(x = var_8103_split_cast_fp16_1)[name = string("op_8103_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_8103_cast_fp16 = mul(x = var_8103_split_cast_fp16_0, y = var_8103_split_1_sigmoid_cast_fp16)[name = string("op_8103_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_923_cast_fp16 = mul(x = var_8103_cast_fp16, y = var_2228_cast_fp16)[name = string("input_923_cast_fp16")];
            string input_925_pad_type_0 = const()[name = string("input_925_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_925_pad_0 = const()[name = string("input_925_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_925_groups_0 = const()[name = string("input_925_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_925_strides_0 = const()[name = string("input_925_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_925_dilations_0 = const()[name = string("input_925_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197806912)))];
            tensor<fp16, [1024]> const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197825408)))];
            tensor<fp16, [1, 1024, 188]> input_927_cast_fp16 = conv(bias = const_86_to_fp16, dilations = input_925_dilations_0, groups = input_925_groups_0, pad = input_925_pad_0, pad_type = input_925_pad_type_0, strides = input_925_strides_0, weight = const_85_to_fp16, x = input_923_cast_fp16)[name = string("input_927_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_137_cast_fp16 = silu(x = input_927_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<int32, [3]> var_8143_perm_0 = const()[name = string("op_8143_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_22_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197827520))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198089728))))[name = string("e_layers_22_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_8143_cast_fp16 = transpose(perm = var_8143_perm_0, x = x_137_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 188, 1024]> var_8145_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_conv_pointwise_conv2_weight_to_fp16_lut, x = var_8143_cast_fp16)[name = string("op_8145_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_929_cast_fp16 = add(x = input_919_cast_fp16, y = var_8145_cast_fp16)[name = string("input_929_cast_fp16")];
            tensor<int32, [1]> input_931_axes_0 = const()[name = string("input_931_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198155328)))];
            tensor<fp16, [1024]> e_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198157440)))];
            fp16 var_8153_to_fp16 = const()[name = string("op_8153_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_931_cast_fp16 = layer_norm(axes = input_931_axes_0, beta = e_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_8153_to_fp16, gamma = e_layers_22_norm_feed_forward2_weight_to_fp16, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_22_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198159552))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199208192))))[name = string("e_layers_22_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_22_feed_forward2_linear1_weight_to_fp16_lut, x = input_931_cast_fp16)[name = string("linear_206_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_935_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_935_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_22_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199470400))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200519040))))[name = string("e_layers_22_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_22_feed_forward2_linear2_weight_to_fp16_lut, x = input_935_cast_fp16)[name = string("linear_207_cast_fp16")];
            fp16 var_8170_to_fp16 = const()[name = string("op_8170_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_8171_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_8170_to_fp16)[name = string("op_8171_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_939_cast_fp16 = add(x = input_929_cast_fp16, y = var_8171_cast_fp16)[name = string("input_939_cast_fp16")];
            tensor<int32, [1]> input_941_axes_0 = const()[name = string("input_941_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_22_norm_out_weight_to_fp16 = const()[name = string("e_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200781248)))];
            tensor<fp16, [1024]> e_layers_22_norm_out_bias_to_fp16 = const()[name = string("e_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200783360)))];
            fp16 var_8175_to_fp16 = const()[name = string("op_8175_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_941_cast_fp16 = layer_norm(axes = input_941_axes_0, beta = e_layers_22_norm_out_bias_to_fp16, epsilon = var_8175_to_fp16, gamma = e_layers_22_norm_out_weight_to_fp16, x = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
            tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("e_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200785472)))];
            tensor<fp16, [1024]> e_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("e_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200787584)))];
            fp16 var_8182_to_fp16 = const()[name = string("op_8182_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = e_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_8182_to_fp16, gamma = e_layers_23_norm_feed_forward1_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_23_feed_forward1_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200789696))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201838336))))[name = string("e_layers_23_feed_forward1_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_23_feed_forward1_linear1_weight_to_fp16_lut, x = input_943_cast_fp16)[name = string("linear_208_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_947_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_947_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_23_feed_forward1_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202100544))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203149184))))[name = string("e_layers_23_feed_forward1_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_feed_forward1_linear2_weight_to_fp16_lut, x = input_947_cast_fp16)[name = string("linear_209_cast_fp16")];
            fp16 var_8199_to_fp16 = const()[name = string("op_8199_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_8200_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_8199_to_fp16)[name = string("op_8200_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_951_cast_fp16 = add(x = input_941_cast_fp16, y = var_8200_cast_fp16)[name = string("input_951_cast_fp16")];
            tensor<int32, [1]> input_953_axes_0 = const()[name = string("input_953_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("e_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203411392)))];
            tensor<fp16, [1024]> e_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("e_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203413504)))];
            fp16 var_8204_to_fp16 = const()[name = string("op_8204_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_953_cast_fp16 = layer_norm(axes = input_953_axes_0, beta = e_layers_23_norm_self_att_bias_to_fp16, epsilon = var_8204_to_fp16, gamma = e_layers_23_norm_self_att_weight_to_fp16, x = input_951_cast_fp16)[name = string("input_953_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_23_self_attn_q_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203415616))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203677824))))[name = string("e_layers_23_self_attn_q_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_self_attn_q_proj_weight_to_fp16_lut, x = input_953_cast_fp16)[name = string("linear_210_cast_fp16")];
            tensor<int32, [4]> var_8217 = const()[name = string("op_8217"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_8218_cast_fp16 = reshape(shape = var_8217, x = linear_210_cast_fp16)[name = string("op_8218_cast_fp16")];
            tensor<int32, [4]> q_perm_0 = const()[name = string("q_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 1024]> e_layers_23_self_attn_k_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203743424))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204005632))))[name = string("e_layers_23_self_attn_k_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_self_attn_k_proj_weight_to_fp16_lut, x = input_953_cast_fp16)[name = string("linear_211_cast_fp16")];
            tensor<int32, [4]> var_8229 = const()[name = string("op_8229"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_8230_cast_fp16 = reshape(shape = var_8229, x = linear_211_cast_fp16)[name = string("op_8230_cast_fp16")];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1024, 1024]> e_layers_23_self_attn_v_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204071232))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204333440))))[name = string("e_layers_23_self_attn_v_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_self_attn_v_proj_weight_to_fp16_lut, x = input_953_cast_fp16)[name = string("linear_212_cast_fp16")];
            tensor<int32, [4]> var_8241 = const()[name = string("op_8241"), val = tensor<int32, [4]>([1, 188, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> var_8242_cast_fp16 = reshape(shape = var_8241, x = linear_212_cast_fp16)[name = string("op_8242_cast_fp16")];
            tensor<int32, [4]> v_perm_0 = const()[name = string("v_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 8, 1, 128]> var_8266_to_fp16 = const()[name = string("op_8266_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204399040)))];
            tensor<fp16, [1, 8, 188, 128]> q_cast_fp16 = transpose(perm = q_perm_0, x = var_8218_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 8, 188, 128]> var_8268_cast_fp16 = add(x = q_cast_fp16, y = var_8266_to_fp16)[name = string("op_8268_cast_fp16")];
            bool input_955_transpose_x_0 = const()[name = string("input_955_transpose_x_0"), val = bool(false)];
            bool input_955_transpose_y_0 = const()[name = string("input_955_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> p_to_fp16 = const()[name = string("p_to_fp16"), val = tensor<fp16, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204401152)))];
            tensor<fp16, [1, 8, 188, 375]> input_955_cast_fp16 = matmul(transpose_x = input_955_transpose_x_0, transpose_y = input_955_transpose_y_0, x = var_8268_cast_fp16, y = p_to_fp16)[name = string("input_955_cast_fp16")];
            tensor<int32, [8]> var_8275_pad_0 = const()[name = string("op_8275_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string var_8275_mode_0 = const()[name = string("op_8275_mode_0"), val = string("constant")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> var_8275_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = var_8275_mode_0, pad = var_8275_pad_0, x = input_955_cast_fp16)[name = string("op_8275_cast_fp16")];
            tensor<int32, [4]> var_8280 = const()[name = string("op_8280"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> var_8281_cast_fp16 = reshape(shape = var_8280, x = var_8275_cast_fp16)[name = string("op_8281_cast_fp16")];
            tensor<int32, [4]> var_8296_begin_0 = const()[name = string("op_8296_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_8296_end_0 = const()[name = string("op_8296_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_8296_end_mask_0 = const()[name = string("op_8296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_8296_cast_fp16 = slice_by_index(begin = var_8296_begin_0, end = var_8296_end_0, end_mask = var_8296_end_mask_0, x = var_8281_cast_fp16)[name = string("op_8296_cast_fp16")];
            tensor<int32, [4]> var_8301 = const()[name = string("op_8301"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> var_8302_cast_fp16 = reshape(shape = var_8301, x = var_8296_cast_fp16)[name = string("op_8302_cast_fp16")];
            tensor<int32, [4]> bd_begin_0 = const()[name = string("bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> bd_end_0 = const()[name = string("bd_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> bd_end_mask_0 = const()[name = string("bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> bd_cast_fp16 = slice_by_index(begin = bd_begin_0, end = bd_end_0, end_mask = bd_end_mask_0, x = var_8302_cast_fp16)[name = string("bd_cast_fp16")];
            tensor<fp16, [1, 8, 1, 128]> var_8313_to_fp16 = const()[name = string("op_8313_to_fp16"), val = tensor<fp16, [1, 8, 1, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205169216)))];
            tensor<fp16, [1, 8, 188, 128]> var_8315_cast_fp16 = add(x = q_cast_fp16, y = var_8313_to_fp16)[name = string("op_8315_cast_fp16")];
            bool ac_transpose_x_0 = const()[name = string("ac_transpose_x_0"), val = bool(false)];
            bool ac_transpose_y_0 = const()[name = string("ac_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 188]> transpose_71_cast_fp16 = transpose(perm = transpose_71_perm_0, x = var_8230_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 8, 188, 188]> ac_cast_fp16 = matmul(transpose_x = ac_transpose_x_0, transpose_y = ac_transpose_y_0, x = var_8315_cast_fp16, y = transpose_71_cast_fp16)[name = string("ac_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_8321_cast_fp16 = add(x = ac_cast_fp16, y = bd_cast_fp16)[name = string("op_8321_cast_fp16")];
            fp16 var_8322_to_fp16 = const()[name = string("op_8322_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> var_8323_cast_fp16 = mul(x = var_8321_cast_fp16, y = var_8322_to_fp16)[name = string("op_8323_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_8325_cast_fp16 = add(x = var_8323_cast_fp16, y = bias_15_cast_fp16)[name = string("op_8325_cast_fp16")];
            int32 var_8326 = const()[name = string("op_8326"), val = int32(-1)];
            tensor<fp16, [1, 8, 188, 188]> w_cast_fp16 = softmax(axis = var_8326, x = var_8325_cast_fp16)[name = string("w_cast_fp16")];
            bool var_8329_transpose_x_0 = const()[name = string("op_8329_transpose_x_0"), val = bool(false)];
            bool var_8329_transpose_y_0 = const()[name = string("op_8329_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> v_cast_fp16 = transpose(perm = v_perm_0, x = var_8242_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 8, 188, 128]> var_8329_cast_fp16 = matmul(transpose_x = var_8329_transpose_x_0, transpose_y = var_8329_transpose_y_0, x = w_cast_fp16, y = v_cast_fp16)[name = string("op_8329_cast_fp16")];
            tensor<int32, [4]> var_8332_perm_0 = const()[name = string("op_8332_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_8336 = const()[name = string("op_8336"), val = tensor<int32, [3]>([1, 188, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_8332_cast_fp16 = transpose(perm = var_8332_perm_0, x = var_8329_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 188, 1024]> input_957_cast_fp16 = reshape(shape = var_8336, x = var_8332_cast_fp16)[name = string("input_957_cast_fp16")];
            tensor<fp16, [1024, 1024]> e_layers_23_self_attn_o_proj_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205171328))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205433536))))[name = string("e_layers_23_self_attn_o_proj_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_self_attn_o_proj_weight_to_fp16_lut, x = input_957_cast_fp16)[name = string("linear_214_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_959_cast_fp16 = add(x = input_951_cast_fp16, y = linear_214_cast_fp16)[name = string("input_959_cast_fp16")];
            tensor<int32, [1]> x_139_axes_0 = const()[name = string("x_139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_23_norm_conv_weight_to_fp16 = const()[name = string("e_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205499136)))];
            tensor<fp16, [1024]> e_layers_23_norm_conv_bias_to_fp16 = const()[name = string("e_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205501248)))];
            fp16 var_8344_to_fp16 = const()[name = string("op_8344_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> x_139_cast_fp16 = layer_norm(axes = x_139_axes_0, beta = e_layers_23_norm_conv_bias_to_fp16, epsilon = var_8344_to_fp16, gamma = e_layers_23_norm_conv_weight_to_fp16, x = input_959_cast_fp16)[name = string("x_139_cast_fp16")];
            tensor<fp16, [2048, 1024]> e_layers_23_conv_pointwise_conv1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [2048, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205503360))), lut = tensor<fp16, [2048, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206027712))))[name = string("e_layers_23_conv_pointwise_conv1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 2048]> var_8358_cast_fp16 = linear(bias = var_2424_bias_0_to_fp16, weight = e_layers_23_conv_pointwise_conv1_weight_to_fp16_lut, x = x_139_cast_fp16)[name = string("op_8358_cast_fp16")];
            tensor<int32, [3]> input_961_perm_0 = const()[name = string("input_961_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_8361_split_num_splits_0 = const()[name = string("op_8361_split_num_splits_0"), val = int32(2)];
            int32 var_8361_split_axis_0 = const()[name = string("op_8361_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 188]> input_961_cast_fp16 = transpose(perm = input_961_perm_0, x = var_8358_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 1024, 188]> var_8361_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> var_8361_split_cast_fp16_1 = split(axis = var_8361_split_axis_0, num_splits = var_8361_split_num_splits_0, x = input_961_cast_fp16)[name = string("op_8361_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_8361_split_1_sigmoid_cast_fp16 = sigmoid(x = var_8361_split_cast_fp16_1)[name = string("op_8361_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> var_8361_cast_fp16 = mul(x = var_8361_split_cast_fp16_0, y = var_8361_split_1_sigmoid_cast_fp16)[name = string("op_8361_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_963_cast_fp16 = mul(x = var_8361_cast_fp16, y = var_2228_cast_fp16)[name = string("input_963_cast_fp16")];
            string input_965_pad_type_0 = const()[name = string("input_965_pad_type_0"), val = string("custom")];
            tensor<int32, [2]> input_965_pad_0 = const()[name = string("input_965_pad_0"), val = tensor<int32, [2]>([4, 4])];
            int32 input_965_groups_0 = const()[name = string("input_965_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_965_strides_0 = const()[name = string("input_965_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> input_965_dilations_0 = const()[name = string("input_965_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206158848)))];
            tensor<fp16, [1024]> const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206177344)))];
            tensor<fp16, [1, 1024, 188]> input_967_cast_fp16 = conv(bias = const_88_to_fp16, dilations = input_965_dilations_0, groups = input_965_groups_0, pad = input_965_pad_0, pad_type = input_965_pad_type_0, strides = input_965_strides_0, weight = const_87_to_fp16, x = input_963_cast_fp16)[name = string("input_967_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_cast_fp16 = silu(x = input_967_cast_fp16)[name = string("x_cast_fp16")];
            tensor<int32, [3]> var_8401_perm_0 = const()[name = string("op_8401_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1024, 1024]> e_layers_23_conv_pointwise_conv2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206179456))), lut = tensor<fp16, [1024, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206441664))))[name = string("e_layers_23_conv_pointwise_conv2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> var_8401_cast_fp16 = transpose(perm = var_8401_perm_0, x = x_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 188, 1024]> var_8403_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_conv_pointwise_conv2_weight_to_fp16_lut, x = var_8401_cast_fp16)[name = string("op_8403_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_969_cast_fp16 = add(x = input_959_cast_fp16, y = var_8403_cast_fp16)[name = string("input_969_cast_fp16")];
            tensor<int32, [1]> input_971_axes_0 = const()[name = string("input_971_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("e_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206507264)))];
            tensor<fp16, [1024]> e_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("e_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206509376)))];
            fp16 var_8411_to_fp16 = const()[name = string("op_8411_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_971_cast_fp16 = layer_norm(axes = input_971_axes_0, beta = e_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_8411_to_fp16, gamma = e_layers_23_norm_feed_forward2_weight_to_fp16, x = input_969_cast_fp16)[name = string("input_971_cast_fp16")];
            tensor<fp16, [4096, 1024]> e_layers_23_feed_forward2_linear1_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206511488))), lut = tensor<fp16, [4096, 8, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207560128))))[name = string("e_layers_23_feed_forward2_linear1_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = e_layers_23_feed_forward2_linear1_weight_to_fp16_lut, x = input_971_cast_fp16)[name = string("linear_215_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_975_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_975_cast_fp16")];
            tensor<fp16, [1024, 4096]> e_layers_23_feed_forward2_linear2_weight_to_fp16_lut = constexpr_lut_to_dense(indices = tensor<uint2, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207822336))), lut = tensor<fp16, [1024, 32, 4, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208870976))))[name = string("e_layers_23_feed_forward2_linear2_weight_to_fp16_lut")];
            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = e_layers_23_feed_forward2_linear2_weight_to_fp16_lut, x = input_975_cast_fp16)[name = string("linear_216_cast_fp16")];
            fp16 var_8428_to_fp16 = const()[name = string("op_8428_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_8429_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_8428_to_fp16)[name = string("op_8429_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_cast_fp16 = add(x = input_969_cast_fp16, y = var_8429_cast_fp16)[name = string("input_cast_fp16")];
            tensor<int32, [1]> h_axes_0 = const()[name = string("h_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> e_layers_23_norm_out_weight_to_fp16 = const()[name = string("e_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209133184)))];
            tensor<fp16, [1024]> e_layers_23_norm_out_bias_to_fp16 = const()[name = string("e_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209135296)))];
            fp16 var_8433_to_fp16 = const()[name = string("op_8433_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> h_cast_fp16 = layer_norm(axes = h_axes_0, beta = e_layers_23_norm_out_bias_to_fp16, epsilon = var_8433_to_fp16, gamma = e_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("h_cast_fp16")];
            tensor<int32, [3]> var_8441_perm_0 = const()[name = string("op_8441_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string var_8441_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_8441_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp16, [1, 1024, 188]> var_8441_cast_fp16 = transpose(perm = var_8441_perm_0, x = h_cast_fp16)[name = string("transpose_0")];
            tensor<fp32, [1, 1024, 188]> encoder = cast(dtype = var_8441_cast_fp16_to_fp32_dtype_0, x = var_8441_cast_fp16)[name = string("cast_0")];
        } -> (encoder);
}